![Page 1: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/1.jpg)
1
ICDCS 2017
IEEE 37th International Conference on Distributed Computing Systems
June 5-8, 2017 Atlanta, USA
Editors
Dr. Kisung Lee Louisiana State University, USA
Dr. Ling Liu Georgia Institute of Technology, USA
![Page 2: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/2.jpg)
2
Table of Contents
Conference Hotel Floor Plan .................................................................................................. 3
Message from the General Chairs and PC Chair .................................................................... 4
Organizing Committee ............................................................................................................ 6
Program Committee ................................................................................................................ 7
Keynotes ............................................................................................................................... 16
Program at a Glance ............................................................................................................. 19
Day 1 – Monday, June 5, 2017 .............................................................................................. 21
Day 2 – Tuesday, June 6, 2017 ............................................................................................. 28
Day 3 – Wednesday, June 7, 2017 ........................................................................................ 39
Day 4 – Thursday, June 8, 2017 ............................................................................................ 49
Research Track Paper Abstracts ......................................................................................... 55
Industry and Experimentation Track Paper Abstracts ........................................................ 73
Applications and Experiences Track Paper Abstracts ........................................................ 76
Vision/Blue Sky Thinking Track Paper Abstracts ................................................................ 88
Short Paper Abstracts .......................................................................................................... 95
Demonstration Track Paper Abstracts .............................................................................. 108
Poster Track Paper Abstracts ............................................................................................ 112
Tutorial Abstracts ............................................................................................................... 119
ADSN 2017 Workshop Abstracts ........................................................................................ 121
BGP 2017 Workshop Abstracts .......................................................................................... 122
CCN-CPS 2017 Workshop Abstracts .................................................................................. 124
HotPOST 2017 Workshop Abstracts .................................................................................. 127
IoTCA 2017 Workshop Abstracts ....................................................................................... 129
JCC 2017 Workshop Abstracts ........................................................................................... 131
PED 2017 Workshop Abstracts ........................................................................................... 134
PSBD 2017 Workshop Abstracts ........................................................................................ 135
WoSC 2017 Workshop Abstracts ........................................................................................ 136
NSF-JST 2017 Workshop Abstracts ................................................................................... 137
Local Information ............................................................................................................... 138
![Page 3: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/3.jpg)
3
Conference Hotel Floor Plan
![Page 4: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/4.jpg)
4
Message from the General Chairs and PC Chair
WewouldliketowelcomeyoutoICDCS2017,the37thIEEEInternationalConferenceonDistributedComputingSystems,beingheldfromJune5toJune8,2017,inAtlanta,Georgia,USA.TheconferenceisheldintheJ.W.MarriottHotel,convenientlylocatedinBuckhead,Atlanta.
Asapremierforumforthepresentationofresearchresultsonabroaderspectrumofdistributedcomputingsystems,the2017ICDCScontinuesthetraditionwithsimilarorganizationalstructure,andintroducessomeinnovations.
Theresearchtrackshavereceivedatotalof531originalsubmissionstotheelevenresearchtracks.Eachpaperwasreviewedbyatleastthreeindependentreviewers,anddiscussedduringtheonlinePCmeetingheldbytrackchairs.Furthermore,thetoprankedpapersfromeachtrackwerefurtherdiscussedduringthefacetofacetrackchairsmeetingheldatGeorgiaInstituteofTechnologyinAtlanta.Atotalof89papershavebeenacceptedbytheresearchtrackswiththeacceptanceratioof16.8%.Wewouldliketoexpressourheartfeltthankstoallofthetrackchairs:MichaelKozuchandAnshulGandhi(CloudandDataCenters),SrinivasAluru(BigDataSystemsandAnalytics),FredDouglis(OSandMiddleware),DougBlough(DistributedFaultTolerantComputing),ElisaBertino(Security,PrivacyandTrust),KarthikSundaresan(WirelessandMobileComputing),KuiRen(DistributedAlgorithms),CanturkIsci(DistributedGreenComputing),ThiemoVoigt(InternetofThingsandCyberPhysicalSystems),WeisongShiandTaoZhang(EdgeandFogComputing),MunindarSingh(SocialandCrowdComputing).
Inaddition,ICDCS2017technicalprogramalsoconsistsofindustrialandexperimentationsessions,applicationsandexperiencessessions,shortpaperssessions,postersanddemossessions,tutorials,andco-locatedworkshops.AnothertwohighlightsofICDCS2017are:(1)thethreekeynotesbydistinguishedspeakers:Dr.C.MohanfromIBMAlmaden,Prof.Dr.TamerOzsufromWaterlooUniversity,andDr.KennethCalvertfromNationalScienceFoundation,and(2)theBlueSkyeIdeasandVisionsessionsfeaturingthought-provokingandforwardlookingblueskyideasandvisionsonbroaderspectrumsoffuturedistributedcomputingsystems.Inaddition,the2017ICDCSmainconferenceisprecededbyeightco-locatedworkshops,describedintheworkshopchairs’message.Theseworkshopsprovideanopportunityforsmallgroupsoflike-mindedresearcherstodiscussareasofinterestandnewideas,andwearepleasedtoseeadiverseecosystemofworkshopsatICDCS.
Thesuccessof2017ICDCSreliedonthecontributionsofagreatteamofvolunteers,includingtheprogramcommitteemembersfromalltracks.Wewouldliketothankallofthemfortheirtime,highqualityservice,andtirelessefforts.Wewouldliketoexpressourspecialappreciationtotheindustryandexperimentationchairs(ScottKlasky,AameekSingh),theblueskyideasandvisiontrackchairs(ManfredHauswirth,ManishParashar),theapplicationsandexperienceschairs(SongqingChen,SangeethaSeshadri),theshortpaperchairs(LaksmishRamaswamy,JianweiYin),andthedemoandposterstrackchairs(GeraldLofstead,BalajiPalanisamy),theworkshopprogramchairs(JoaoE.Ferreira,TeruoHigashino),tutorialchairs(JamesCaverlee,VaidySunderam),panelchairs(DimitriosGeorgakopoulos,YiPan,YinglongXia),publicationchair(KisungLee),publicitychair(AibekMusaev,EmreYigitoglu),localarrangementchairs(AibekMusaev,JoshuaKimball,WeiZhou),andfinance/registrationchair(CarrieStein).
Finally,onbehalfofthe2017ICDCSorganizationandprogramcommittee,wewouldliketoexpressoursinceregratitudetoeveryonewhohascontributedtotheconference,especiallytheauthorsandtheparticipants.WewouldliketothankXiaodongZhang(ICDCSSteeringCommitteeChair)andCheng-ZhongXu(chairofIEEEComputerSocietyTechnicalCommitteeonDistributedProcessing)fortheirsupport,adviceandtrustingustoorganizetheconference.Theyalsospearheadedthesuccessfulefforttoobtainanddisbursefundingtosponsorthestudenttravelawards.Inaddition,wewouldliketoacknowledgethesponsorshipsupportbytheNationalScienceFoundation(NSF/USA)andJapanScienceandTechnologyAgency(JST/Japan)forthe1stUS-JapanWorkshoponEnablingGlobalCollaborationsinBigDataResearch.
![Page 5: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/5.jpg)
5
WewishyouanenjoyableandproductiveconferenceandapleasantstayinAtlanta.
ProgramChair:
LingLiu,GeorgiaInstituteofTechnology,USA
GeneralChairs:
CaltonPu,GeorgiaInstituteofTechnology,USA
MasaruKitsuregawa,NIIandU.Tokyo,Japan
KarlAberer,EPFL,Switzerland
![Page 6: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/6.jpg)
6
Organizing Committee
General Co-Chairs Calton Pu, Georgia Institute of Technology, USA Masaru Kitsuregawa, University of Tokyo, Japan Karl Aberer, EPFL, Switzerland
Program Chair
Ling Liu, Georgia Institute of Technology, USA Workshops Program Chairs
Joao E. Ferreira, University of Sao Paulo, Brazil Teruo Higashino, University of Osaka, Japan
Tutorial Chairs
James Caverlee, Texas A&M University, USA Vaidy Sunderam, Emory University, USA
Panel Chairs
Dimitrios Georgakopoulos, Swinburne University, Australia Yi Pan, Georgia State University, USA Yinglong Xia, Huawei Research America, USA
Short Papers Track Chairs
Laksmish Ramaswamy, UGA, USA Jianwei Yin, Zhejiang University, PR China
Demonstration and Posters Track Chairs
Gerald F Lofstead, Sandia National Laboratories, USA Balaji Palanisamy, University of Pittsburg, USA
Publication Chair
Kisung Lee, Louisiana State University, USA Workshops Publication Chair
Aibek Musaev, University of Alabama, USA Publicity Chair, Local Arrangement Chair, Webmaster
Aibek Musaev, University of Alabama, USA Publicity Co-Chair
Emre Yigitoglu, Georgia Institute of Technology, USA Steering Committee Chair
Xiaodong Zhang, Ohio State University, USA Finance Manager (non-volunteer, reporting to General Chairs)
Carrie Stein, Ohio State University, USA
![Page 7: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/7.jpg)
7
Program Committee
Program Chair
Ling Liu, Georgia Institute of Technology, USA Program Vice Chairs and TPC Members (Research Tracks) § Cloud Computing and Data Centers
Co-Chair: Michael A. Kozuch, Intel, USA Co-Chair: Anshul Gandhi, Stony Brook University, USA § George Amvrosiadis, Carnegie Mellon University § Ganesh Ananthanarayanan, Microsoft Research § Danilo Ardagna, Politecnico di Milano § Alvin AuYoung, Uber § Bharath Balasubramanian, ATT Labs Research § Sonia Ben Mokhtar, LIRIS CNRS § Sara Bouchenak, INSA, Lyon § Ali R. Butt, Virginia Tech § Irina Calciu, VMware Research Group § Maria Carla Calzarossa, Universita di Pavia § Giovanni Toffetti Carughi, ZHAW § Giuliano Casale, Imperial College London § Claris Castillo, RENCI@UNC Chapel Hill § Lydia Y. Chen, IBM Research Zurich Lab § Keke Chen, Wright State University § Byung-Gon Chun, Seoul National University § Erik Elmroth, Umeå University and Elastisys § Renato Figueiredo, University of Florida § Sahan Gamage, VMware § Christos Gkantsidis, Microsoft Research § Chris Gniady, University of Arizona § Kartik Gopalan, Binghamton University § Daniel Grosu, Wayne State University § Xiaohui Gu, North Carolina State University § Haryadi S. Gunawi, University of Chicago § Yu Hua, Huazhong University of Science and Technology § Alexandru Iosup, Delft University of Technology § Evangelia Kalyvianaki, City, University of London § Eric Keller, University of Colorado, Boulder § Steve Ko, SUNY Buffalo § Donald Kossmann, Microsoft Research Redmond & ETH Zurich § Diwakar Krishnamurthy, University of Calgary § Purushottam Kulkarni, Indian Institute of Technology, Bombay § John Lange, University of Pittsburgh § Marin Litoiu, York University § Zhenhua Liu, Stony Brook University § Fangming Liu, Huazhong University of Science and Technology § Amiya Maji, Purdue University § Lena Mashayekhy, University of Delaware § Thu Nguyen, Rutgers University § Balaji Palanisamy, University of Pittsburgh
![Page 8: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/8.jpg)
8
§ Donald E. Porter, The University of North Carolina at Chapel Hill § Chen Qian, University of California Santa Cruz § Ioan Raicu, Illinois Institute of Technology § Shriram Rajagopalan, IBM Watson Research § Shaolei Ren, University of California, Riverside § Henry Robinson, Cloudera § Stefan Schmid, Aalborg University & TU Berlin § Haiying Shen, University of Virginia § Evgenia Smirni, College of William and Mary § Yuzhe Tang, Syracuse University § Devesh Tiwari, ORNL § Bhuvan Urgaonkar, Pennsylvania State University § Xiaorui Wang, The Ohio State University § Anduo Wang, Temple University § Jianguo Yao, Shanghai Jiao Tong University § Ming Zhao, Arizona State University § Xiaobo Zhou, University of Colorado, Colorado Springs § Xiaoyun Zhu, Futurewei Technologies, Inc.
§ Distributed Big Data Systems & Analytics Chair: Srinivas Aluru, Georgia Institute of Technology, USA § Gagan Agrawal, Ohio State University § Rajesh Bordawekar, IBM T. J. Watson Research Center § Polo Chau, Georgia Institute of Technology § Alin Dobra, University of Florida § Vijay Gadepally, Massachusetts Institute of Technology § Bugra Gedik, Bilkent University, Turkey § Tony Xiaohua Hu, Drexel University § Hans-Arno Jacobsen, University of Toronto § George Karypis, University of Minnesota § Tevfik Kosar, State University of New York at Buffalo § Zhenhui (Jessie) Li, Pennsylvania State University § Ron Oldfield, Sandia National Laboratories § Srinivasan Parthasarathy, Ohio State University § Peter Pietzuch, Imperial College § Vijay Raghavan, Louisiana State University § Yogesh Simmhan, Indian Institute of Science § Nesime Tatbul, Intel Labs and Massachusetts Institute of Technology § Ying Xie, Kennesaw State University § Jianfeng Zhan, Chinese Academy of Sciences § Justin Z. Zhan, University of Nevada Las Vegas
§ Distributed Operating Systems & Middleware Chair: Fred Douglis, Dell EMC, USA § Paolo Bellavista, University of Bologna, Italy § Surendar Chandra, Datrium, USA § Yuan Chen, Hewlett Packard Labs, USA § Geoffrey Fox, Indiana University, USA § Indranil Gupta, University of Illinois at Urbana-Champaign, USA § Yu Hua, Huazhong University of Science and Technology, China
![Page 9: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/9.jpg)
9
§ Hong Jiang, University of Texas at Arlington, USA § Vana Kalogeraki, Athens Institute of Economics and Business, Greece § Fabio Kon, University of São Paulo, Brazil § Xiaosong Ma, Qatar Computing Research Institute, Qatar § Rajesh Panta, AT&T Labs – Research, USA § Luís Rodrigues, IST, ULisboa, Portugal § Liuba Shrira, Brandeis University, USA § Francois Taiani, University of Rennes 1, IRISA / Inria, France § Maarten Van Steen, University of Twente, Netherlands § Chitra Venkatramani, Google, USA § Zhen Xiao, Peking Univ, China § Gala Yadgar, Technion, Israel § Ming Zhao, Arizona State University, USA
§ Distributed Algorithms and Theory Chair: Kui Ren, SUNY, Buffalo, USA § Jiannong Cao, Hong Kong Polytechnic University § Xiuzhen Cheng, The George Washington University § Ding-Zhu Du, University of Texas at Dallas § Michael Fang, University of Florida § Vincent Gramoli, University of Sydney § Xiaohua Jia, City University of Hong Kong § Murat Kantarcioglu, University of Texas at Dallas § Nei Kato, Tohoku University § Sastry Kompella, Naval Research Lab § Sam Kwong, City University of Hong Kong § Baochun Li, University of Toronto § Xiangyang Li, University of Science and Technology of China § Phone Lin, National Taiwan University § Benyuan Liu, University of Massachusetts Lowell § Jiangchuan Liu, Simon Fraser University § Xue Liu, McGill University § Tommaso Melodia, Northeastern University § Aziz Mohaisen, SUNY Buffalo § Guevara Noubir, Northeastern University § Francesco Pasquale, Sapienza Università di Roma § Sherman Shen, University of Waterloo § Paul Spirakis, RA.CTI and University of Patras § Anna Squicciarini, The Pennsylvania State University § Lu Su, SUNY Buffalo § My Thai, University of Florida § Peng-Jun Wan, Illinois Institiute of Technology § Jie Wu, Temple University § Nicholas Zhang, Huawei Technologies § Sheng Zhong, SUNY Buffalo
§ Distributed Fault Tolerance & Dependability Chair: Doug Blough, Georgia Institute of Technology, USA § Emmanuelle Anceaume, IRISA, France § Saurabh Bagchi, Purdue University, USA
![Page 10: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/10.jpg)
10
§ Antonio Casimiro, University of Lisbon, Portugal § Xavier Defago, Tokyo Institute of Technology, Japan § Felicita di Giandomenico, ISTI-CNR, Italy § Elias Duarte, Federal University of Parana, Brazil § Christof Fetzer, Technical University of Dresden, Germany § Roy Friedman, Technion, Israel § Patrick Lee, Chinese University of Hong Kong, China § Michael Lyu, Chinese University of Hong Kong, China § Cristina Nita-Rotaru, Northeastern University, USA § Fernando Pedone, University of Lugano, Switzerland § Sebastiano Peluso, Virginia Tech, USA § Ravi Prakash, University of Texas at Dallas, USA § Tatsuhiro Tsuchiya, Osaka University, Japan § Ymir Vigfusson, Emory University, USA
§ Distributed Green Computing & Energy Management Chair: Canturk Isci, IBM T.J. Watson, USA § Niklas Carlsson, Linköping University, Sweden § Ayse Coskun, Boston University § Yuanxiong (Richard) Guo, Oklahoma State University § David Irwin, University of Massachusetts Amherst § Prabhakar Kudva, IBM TJ Watson Research Center § Fangming Liu, Huazhong University of Science & Technology § Vijay Janapa Reddi, University of Texas at Austin § Suzanne McIntosh, Cloudera & NYU § Shaolei Ren, University of California, Riverside § Tajana Simunic Rosing, University of California, San Diego § Anand Sivasubramaniam, Pennsylvania State University § Christopher Stewart, Ohio State University § Devesh Tiwari, Oak Ridge National Laboratory § Di Wang, Microsoft Research, Redmond § Carole-Jean Wu, Arizona State University
§ Internet of Things & Cyber-Physical Systems Chair: Thiemo Voigt, SICS and Uppsala University, Sweden § Tarek Abdelzaher, University of Illinois at Urbana Champaign, USA § Carlo Boano, TU Graz, Austria § Qing Cao, University of Tennessee, USA § Octav Chipara, University of Iowa, USA § Wan Du, Nanyang Technological University, Singapore § Simon Duquennoy, Inria, France, and SICS, Sweden § Yuan He, Tsinghua University, China § Wen Hu, University of New South Wales, Australia § JeongGil Ko, Ajou University, South Korea § Linghe Kong, Shanghai Jiao Tong University, China § Mike Chieh-Jan Liang, Microsoft Research Asia § Margaret Loper, Georgia Institute of Technology, USA § Pedro Marron, University of Duisburg-Essen, Germany § Luca Mottola, Politecnico di Milano, Italy and SICS, Sweden § Edith Ngai, Uppsala University, Sweden
![Page 11: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/11.jpg)
11
§ Gian Pietro Picco, University of Trento, Italy § Mo Sha, SUNY Binghamton, USA § David Smith, Data61, CSIRO, Australia § Eduardo Tovar, Polytechnic Institute of Porto, Portugal § Chenshu Wu, Tsinghua University, China § I-Ling Yen, University of Texas at Dallas, USA § Haibo Zhang, University of Otago, New Zealand
§ Mobile and Wireless Network Computing Chair: Karthik Sundaresan, NEC Labs Princeton, USA § Ehsan Aryafar, Intel, USA § Zhipeng Cai, Georgia State Univ § Mehmet Can Vuran, Univ of Nebraska Lincoln § Guohong Cao, Penn State Univ § Aveek Dutta, Univ at Albany § Monisha Ghosh, Univ of Chicago § Tian He, Univ of Minnesota § Ravi Kokku, IBM research § Dimitrios Koutsonikolas, Univ of Buffalo § Sandeep Kulkarni, Michigan State Univ § Mo Li, NTU § Shiwen Mao, Auburn Univ § Joerg Ott, Aalto University § Neal Patwari, Univ of Utah § Feng Qian, Indiana University § Sampath Rangarajan, NEC Labs America § Kay Romer, ETH Zurich § Theodoros Salonidis, IBM research § Thrasyvoulos Spyropoulos, Eurocom § Lu Su, Univ of Buffalo § Nalini Venkatasubramanian, Univ of California, Irvine § Xinbing Wang, Shanghai Jiao Tong Univ § Jie Wu, Temple Univ § Jie Xiong, Singapore Management University § Jongwon Yoon, Hanyang university, Korea § Moustafa A. Youssef, Egypt-Japan University of Science and Technology (E-JUST)
§ Edge and Fog Computing Co-Chair: Weisong Shi, Wayne State University, USA Co-Chair: Tao Zhang, Cisco, USA § Gary Chan, The Hong Kong University of Science and Technology § Dilma Da Silva, Texas A&M University § Schahram Dustdar, TU Wien § Sangtae Ha, University of Colorado at Boulder § Robert Hall, AT&T Labs Research § Bhaskar Krishnamachari, USC § Tian Lan, GWU § Martin Lehofer, Siemens VAI Metals Technologies GmbH § Qun Li, College of William and Mary § Michael Rabinovich, Case Western Reserve University
![Page 12: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/12.jpg)
12
§ Eve Schooler, Intel Research § Andrew Weinert, MIT § Fan Ye, Stony Brook University § Junshan Zhang, Arizona State University § Yanyong Zhang, Rutgers University
§ Security, Privacy, Trust in Distributed Systems Chair: Elisa Bertino, Purdue University, USA § Ehab Al-Shaer, UNC Charlotte § Salman A Baset, IBM Research § Nilanjan Banerjee, UMBC § Barbara Carminati, University of Insubria, Italy § Aniello Castiglione, University of Salerno, Italy § Dario Catalano, University of Catania, Italy § Yan Chen, Northwestern University § Lucas Davi, University of Duisburg-Essen, Germany § Robert Deng, Singapore Management University § Elena Ferrari, University of Insubria, Italy § Aurelien Francillon, Eurecom § Michael Franz, University of California at Irvine § Gabriel Ghinita, UMass at Boston § Guofei Gu, Texas A&M § Xinyi Huang, Fujian Normal University § Aniket Kate, Purdue University § Florian Kerschbaum, SAP § Taesoo Kim, Georgia Institute of Technology § Zhenkai Liang, National University of Singapore § Dan Lin, Missouri University of Science and Technology § Alex Liu, Michigan State University § Surya Nepal, CSIRO, Australia § Kui Ren, SUNY Buffalo § Seung-Yun Seo, Korea University § Anna Squicciarini, Pennsylvania State University § Cong Wang, Hong Kong City University § Danfeng Yao, Virginia Tech § Xun Yi, RMIT, Australia § Ting Yu, Qatar Computing Research Institute
§ Social Networks and Crowdsourcing Chair: Munindar P. Singh, NCSU, USA § Alessandro Bozzon, Delft University of Technology, Netherlands § James Caverlee, Texas A&M University, USA § Peng Cui, Tsinghua University, China § Raghu Ganti, IBM T J Watson Research Center, USA § Chung-Wei Hang, IBM, USA § Catholijn Jonker, Delft University of Technology, Netherlands § Anup Kalia, IBM T. J. Watson Research Center, USA § Vana Kalogeraki, Athens University of Economics and Business, Greece § Sharad Mehrotra, University of California, Irvine, USA § Pradeep Kumar Murukannaiah, Rochester Institute of Technology, USA
![Page 13: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/13.jpg)
13
§ Mirco Musolesi, University College London, UK § Nishanth Sastry, King’s College London, UK § Murat Sensoy, Ozyegin University, Turkey § Erez Shmueli, Tel-Aviv University, Turkey § Mudhakar Srivatsa, IBM T.J. Watson Research Center, USA § Guangchao Yuan, Microsoft, USA § Zhe Zhang, IBM Watson, USA
Special Tracks TPC Chairs and TPC Members
§ Industry and Experimentation
Co-Chair: Scott Klasky, Oak Ridge National Lab. USA Co-Chair: Aameek Singh, IBM T.J. Watson § Dorian Arnold, UNM, USA § Rong Chang, IBM, USA § Carlos Costa, IBM, USA § Alfredo Cuzzocrea, University of Trieste, Italy § Raghu Ganti, IBM, USA § Garth Gibson, CMU, USA § Tahsin Kurc, SBU, USA § Qing Liu, NJIT, USA § Xuanzhe Liu, Peking University, China § Heiko Ludwig, IBM, USA § Carlos Maltzahn, UCSC, USA § Nagapramod Mandagere, IBM, USA § Manish Parashar, Rutgers, USA § Yang Song, IBM, USA § Matthew Wolf, Oak Ridge National Lab., USA § Pengcheng Xiong, Hortonworks, USA § Pingpeng Yuan, Huazhong University of Science and Technology, China § Fan Zhang, IBM, USA § Yiming Zhang, National University of Defense Technology, China § Xiaomin Zhu, National University of Defense University, China
§ Applications and Experiences
Co-Chair: Songqing Chen, George Mason University, USA Co-Chair: Sangeetha Seshadri, IBM Almaden, USA § Kun Bai, IBM TJ Watson Research Center, USA § Bhuvan Bamba, Oracle, USA § Anu Bourgeois, Georgia State University, USA § Umesh Deshpande, IBM, USA § Xiaoning Ding, New Jersey Institute of Technology, USA § Vijay Gopalakrishnan, AT&T Labs – Research, USA § Yang Guo, NIST, USA § Vana Kalogeraki, Athens University of Economics and Business, Greece § Kisung Lee, Louisiana State University, USA § Qi Li, Tsinghua University, China § Fei Li, George Mason University, USA § Baochun Li, University of Toronto, Canada § Yao Liu, SUNY Binghamton, USA
![Page 14: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/14.jpg)
14
§ Aziz Mohaisen, SUNY Buffalo, USA § Shannon Quinn, University of Georgia, USA § Lukas Rupprecht, Imperial College London, UK § Aki-Hiro Sato, Kyoto University, Japan § Viswanathan Swaminathan, Adobe, USA § Masashi Toyoda, Univ Tokyo, Japan § Qingyang Wang, Louisiana State University, USA § Ting Wang, Lehigh University, USA § Li Xiao, Michigan State University, USA § Weijun Xiao, Virginia Commonwealth University, USA § Jianwei Yin, Zhejiang University, China § Gong Zhang, Oracle Corporation, USA § Rui Zhang, Institute of Information Engineering, Chinese Academy of Sciences, China § Yanqing Zhang, Georgia State University, USA § Zhao Zhang, University of Illinois at Chicago, USA § Xuan Zhou, Renmin University of China, China
§ Vision / Blue Sky Thinking Co-Chair: Manfred Hauswirth, TU Berlin, Germany Co-Chair: Manish Parashar, Rutgers University, USA § Ozalp Babaoglu, University of Bologna, Italy § Umesh Bellur, IIT Bombay, India § Jesus Carretero, Universidad Carlos III de Madrid, Spain § Frederic Desprez, INRIA, France § Dieter Kranzlmüller, Ludwig-Maximilians-Universitaet Muenchen, Germany § Christine Morin, INRIA, France § Thu Nguyen, Rutgers University, USA § Vladimir Vlassov, Royal Institute of Technology (KTH), Sweden
Short Papers Track
Co-Chair: Laksmish Ramaswamy, UGA, USA Co-Chair: Jianwei Yin, Zhejiang University, PR China § Amir Abdolrashidi, The University of Georgia, USA § Cao Bin, College of Computer Science and Technology, Zhejiang University of Technology, China § Vinay Boddula, University of Georgia, USA § Arash Fard, Hewlett Packard Enterprise, USA § Rong Ge, Clemson University, USA § Takahiro Hara, Graduate School of Information Science and Technology, Osaka University, Japan § Kyu Lee, University of Georgia, USA § Wenjia Li, New York Institute of Technology, USA § Xinjian Lu, East China University of Science and Technology, China § Shuai Ma, Beihang University, China § Rohit Mullangi, University of Georgia, USA § Aibek Musaev, University of Alabama, USA § Surya Nepal, CSIRO, Australia § Balaji Palanisamy, University of Pittsburgh, USA § Shannon Quinn, University of Georgia, USA § Michael Sheng, The University of Adelaide, Australia § Varun Soundararajan, Google, USA § Anna Squicciarini, The Pennsylvania State University, USA
![Page 15: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/15.jpg)
15
§ Yueshen Xu, Xidian University, China § Yun Yang, Swinburne University of Technology, Australia § Emre Yigitoglu, Georgia Institute of Technology, USA
Demonstration and Posters Track
Co-Chair: Gerald F Lofstead, Sandia Nat. Lab., USA Co-Chair: Balaji Palanisamy, University of Pittsburg, USA § Nathalie Baracaldo, IBM Research, USA § Suren Byna, Lawrence Berkeley National Laboratory, USA § Toni Collis, EPCC, UK § Liting Hu, Florida International University, USA § Suzanne McIntosh, NYU Courant Institute, and Cloudera Inc., USA § Devesh Tiwari, ORNL, USA § Ming Zhao, Arizona State University, USA
![Page 16: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/16.jpg)
16
Keynotes
Keynote 1: C. Mohan (IBM Research)
NewErainDistributedComputingwithBlockchainsandDatabases
Aneweraisemergingintheworldofdistributedcomputingwiththegrowingpopularityofblockchains(shared,replicatedanddistributedledgers)andtheassociateddatabasesasawayofintegratinginter-organizationalwork.Originally,theconceptofadistributedledgerwasinventedastheunderlyingtechnologyofthecryptocurrencyBitcoin.Buttheadoptionandfurtheradaptationofitforuseinthecommercialorpermissionedenvironmentsiswhatisofutmostinteresttomeandhencewillbethefocusofthiskeynote.ComputercompanieslikeIBMandMicrosoft,andmanykeyplayersindifferentverticalindustrysegmentshaverecognizedtheapplicabilityofblockchainsinenvironmentsotherthancryptocurrencies.IBMdidsomepioneeringworkbyarchitectingandimplementingFabric,andthenopensourcingit.NowFabricisbeingenhancedviatheHyperledgerConsortiumaspartofTheLinuxFoundation.AfewoftheothereffortsincludeEnterpriseEthereum,R3CordaandBigchainDB.
Whilethereisnostandardintheblockchainspacecurrently,alltheongoingeffortsinvolvesomecombinationofdatabase,transaction,encryption,consensusandotherdistributedsystemstechnologies.Someoftheapplicationareasinwhichblockchainpilotsarebeingcarriedoutare:smartcontracts,supplychainmanagement,knowyourcustomer,derivativesprocessingandprovenancemanagement.Inthistalk,Iwillsurveysomeoftheongoingblockchainprojectswithrespecttotheirarchitecturesingeneralandtheirapproachestosomespecifictechnicalareas.Iwillfocusonhowthefunctionalityoftraditionalandmoderndatastoresarebeingutilizedornotutilizedinthedifferentblockchainprojects.Iwillalsodistinguishhowtraditionaldistributeddatabasemanagementsystemshavehandledreplicationandhowblockchainsystemsdoit.Sincemostoftheblockchaineffortsarestillinanascentstate,thetimeisrightfordatabaseandotherdistributedsystemsresearchersandpractitionerstogetmoredeeplyinvolvedtofocusonthenumerousopenproblems.
Bio:Dr.C.MohanhasbeenanIBMresearcherfor35yearsinthedatabasearea,impactingnumerousIBMandnon-IBMproducts,theresearchandacademiccommunities,andstandards,especiallywithhisinventionoftheARIESfamilyofdatabaselockingandrecoveryalgorithms,andthePresumedAbortcommitprotocol.ThisIBM(1997),andACM/IEEE(2002)FellowhasalsoservedastheIBMIndiaChiefScientistfor3years(2006-2009).InadditiontoreceivingtheACMSIGMODInnovationAward(1996),theVLDB10YearBestPaperAward(1999)andnumerousIBMawards,MohanwaselectedtotheUSandIndianNationalAcademiesofEngineering(2009),andwasnamedanIBMMasterInventor(1997).ThisDistinguishedAlumnusofIITMadras(1977)receivedhisPhDattheUniversityofTexasatAustin(1981).Heisaninventorof50patents.HeiscurrentlyfocusedonBigData,HTAPandBlockchaintechnologies.In2016,hewasnamedaDistinguishedVisitingProfessorofChina’sprestigiousTsinghuaUniversity.HehasservedontheadvisoryboardofIEEESpectrum,andonnumerousconferenceandjournalboards.MohanisafrequentspeakerinNorthAmerica,EuropeandIndia,andhasgiventalksin40countries.Heisveryactiveonsocialmediaandhasahugenetworkoffollowers.MoreinformationcouldbefoundintheWikipediapageathttp://bit.ly/CMwIkP
![Page 17: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/17.jpg)
17
Keynote 2: M. Tamer Özsu (University of Waterloo)
WebDataManagementintheRDFAge
Webdatamanagementhasbeenatopicofinterestformanyyearsduringwhichanumberofdifferentmodellingapproacheshavebeentried.ThelatestinthisapproachesistouseRDF(ResourceDescriptionFramework),whichseemstoproviderealopportunityforqueryingatleastsomeofthewebdatasystematically.RDFhasbeenproposedbytheWorldWideWebConsortium(W3C)formodelingWebobjectsaspartofdevelopingthe“semanticweb”.W3ChasalsoproposedSPARQLasthequerylanguageforaccessingRDFdatarepositories.ThepublicationofLinkedOpenData(LOD)ontheWebhasgainedtremendousmomentumoverthelastnumberofyears,andthisprovidesanewopportunitytoaccomplishwebdataintegration.AnumberofapproacheshavebeenproposedforrunningSPARQLqueriesoverRDF-encodedWebdata:datawarehousing,SPARQLfederation,andlivelinkedqueryexecution.Inthistalk,IwillreviewtheseapproacheswithparticularemphasisonsomeofourresearchwithinthecontextofgStoreproject(jointprojectwithProf.LeiZouofPekingUniversityandProf.LeiChenofHongKongUniversityofScienceandTechnology),chameleon-dbproject(jointworkwithGünesAluç,Dr.OlafHartig,andProf.KhuzaimaDaudjeeofUniversityofWaterloo),andlivelinkedqueryexecution(jointworkwithDr.OlafHartig).
Bio:M.TamerÖzsuisProfessorofComputerScienceattheDavidR.CheritonSchoolofComputerScience.Hisresearchisindatamanagementfocusingonlarge-scaledatadistributionandmanagementofnon-traditionaldata.HeisaFellowoftheRoyalSocietyofCanada,oftheAssociationforComputingMachinery(ACM),andoftheInstituteofElectricalandElectronicsEngineers(IEEE).HeisanelectedmemberoftheScienceAcademyofTurkey,andamemberofSigmaXiandAmericanAssociationfortheAdvancementofScience(AAAS).HispublicationsincludethebookPrinciplesofDistributedDatabaseSystems(withPatrickValduriez),whichisnowinitsthirdedition.Hehasalsoedited,withLingLiu,theEncyclopediaofDatabaseSystems,whichisnowinitssecondedition.HewastheFoundingSeriesEditorofSynthesisLecturesonDataManagement(Morgan&Claypool),andisnowtheEditor-in-ChiefofACMBooks.Heservesontheeditorialboardsofthreejournals,andtwobookSeries.Prof.Dr.M.TamerÖzsuwasawardedtheACMSIGMODTest-of-TimeAwardin2015,ACMSIGMODContributionsAwardin2008andtheOhioStateUniversityCollegeofEngineeringDistinguishedAlumnusAwardin2008.
![Page 18: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/18.jpg)
18
Keynote 3: Kenneth Calvert (National Science Foundation)
APerspectiveonDistributedComputingSystems:Past,PresentandFuture
ThiskeynotetalkwillofferabroadviewofdistributedcomputingfromtheperspectiveofpastandpresentNSFinvestments.Itwillincludealooktowardfuturechallengesandopportunitiesfacingthedistributedcomputingresearchcommunity.
Bio:Dr.KennethCalvertisDivisionDirectorforComputerandNetworkSystemsintheComputerandInformationScienceandEngineering(CISE)DirectorateattheNationalScienceFoundation.HeisonrotationfromtheUniversityofKentucky,whereheisGartnerGroupProfessorinNetworkEngineeringintheDepartmentofComputerScience.Hisresearchdealswiththedesignandimplementationofadvancednetworkprotocolsandservices,withparticularinterestinroutingandincentivesinfuturenetworkarchitectures.HereceivedhisPh.D.incomputersciencefromtheUniversityofTexasatAustin.HeholdsaM.S.incomputersciencefromStanfordUniversityandaB.S.incomputerscienceandengineeringfromtheMassachusettsInstituteofTechnology.PriortohisappointmentattheUniversityofKentucky,hewasaMemberoftheTechnicalStaffatBellLaboratoriesinHolmdel,NJ,andservedonthefacultyintheCollegeofComputingattheGeorgiaInstituteofTechnology.HeisanIEEEFellowandamemberoftheACM.
![Page 19: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/19.jpg)
19
Program at a Glance
Day1:Monday,June5,2017
Time Track1(SalonII)
Track2(SalonIV)
Track3(SalonVI)
Track4(Atlanta)
Track5(Columbia)
Track6(Savannah)
Track7(Charleston1)
7:00-8:00 ContinentalBreakfast(Foyer)8:00-9:30
Workshop:CCN-CPS Workshop:
HotPOSTWorkshop:
PSBDWorkshop:
JCC
Workshop:NSF-JSTDay1
Tutorial1:Serverless
Programming(Functionasa
Service)9:30-10:00 CoffeeBreak(Foyer)10:00-12:00
Workshop:CCN-CPS
Workshop:ADSN
Workshop:HotPOST
Workshop:PSBD
Workshop:JCC
Workshop:NSF-JSTDay1
Tutorial1:Serverless
Programming(Functionasa
Service)12:00-13:30 Lunch(PhoenixBallroom)13:30-15:30 Workshop:
CCN-CPSWorkshop:
ADSNWorkshop:PED-BGP
Workshop:IoTCA
Workshop:WoSC
Workshop:NSF-JSTDay1
15:30-16:00 CoffeeBreak(Foyer)
16:00-17:00Workshop:CCN-CPS
Workshop:ADSN
Workshop:PED-BGP
Workshop:IoTCA
Workshop:WoSC
Workshop:NSF-JSTDay1
17:00-18:00 Workshop:
PED-BGP
Day2:Tuesday,June6,2017
Time Track1(SalonI)
Track2(SalonII)
Track3(Atlanta)
Track4(Columbia)
Track5(Nashville)
Track6(Charleston1)
7:00-8:00 ContinentalBreakfast(Foyer)8:00-8:30 ConferenceOpening(PhoenixBallroom)8:30-9:30 Keynote1(PhoenixBallroom)9:30-10:00 CoffeeBreak(Foyer)
10:00-12:00Research1:
DistributedFaultToleranceandDependability
Research2:DistributedOperating
SystemsandMiddleware
Vision1:InternetofThings,SmartCitiesandCyber-Physical
Systems
Application1:Security,Privacy,TrustinDistributedSystems
Application2:SocialNetworksand
Crowdsourcing
Industry1:CloudDataCentersandPerformance
12:00-13:30 Lunch(PhoenixBallroom)
13:30-15:30Research3:
SecurityandPrivacyinDistributedSystemsI
Research4:CloudComputingandDataCenterSystems
Vision2:FutureNetworking
andCyberinfrastructure
Application3:InternetofThings,SmartCities,andCyber-Physical
Systems
Application4:Mobile,Wireless,and
EdgeComputing
ShortPaper1:DistributedOperatingSystems,Middleware,
andAlgorithms
15:30-16:00 CoffeeBreak(Foyer)
16:00-18:00Research5:EdgeandFogComputing
Research6:DistributedGreen
ComputingandEnergyManagement
Vision3:NextGenerationCloudandEdge
Services
ShortPaper2:CloudandDataCenterSystemsandNetworks
ShortPaper3:InternetofThings,SmartCities,andCyber-Physical
Systems
Poster1–4:16:00-17:00Lighteningtalks
(Charleston1&2)17:00-19:00
Display(PhoenixBallroom)
10:00-18:00 The1stUS-JapanWorkshoponCollaborativeGlobalResearchonApplyingInformationTechnologyDay2(Savannah)
18:00-20:00 Reception(PhoenixBallroom)
![Page 20: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/20.jpg)
20
Day3:Wednesday,June7,2017
Time Track1(SalonI)
Track2(SalonII)
Track3(Atlanta)
Track4(Columbia)
Track5(Nashville)
Track6(Charleston1)
7:30-8:30 ContinentalBreakfast(Foyer)8:30-9:30 Keynote2(PhoenixBallroom)9:30-10:00 CoffeeBreak(Foyer)
10:00-12:00
Research7:InternetofThings,SmartCities,andCyber-Physical
Systems
Research8:MobileandWirelessComputingSystemsI
Vision4:SecurityandTrustin
FutureSystems
Application5:CloudComputingandDataCenterSystems
Application6:BigDataSystemsandDistributedDataManagementand
Analytics
Industry2:MobileComputingand
InternetofThings
12:00-13:30 ConferenceLuncheon(PhoenixBallroom)
13:30-15:30Research9:
DistributedBigDataSystems
Research10:DistributedAlgorithms
andTheoryI
Vision5:FutureDistributed
Systems
Application7:Distributed
MiddlewareSystems
Application8:DistributedSystemsandOptimizations
ShortPaper4:Mobile,Wireless,Edge,andCrowd
Computing15:30-16:00 CoffeeBreak(Foyer)
16:00-18:00Research11:
SecurityandPrivacyinDistributedSystemsII
Research12:CloudComputingand
DistributedDataAnalytics
Vision6:InnovationinBigData
Systems
Research13:DistributedAlgorithms
andTheoryII
ShortPaper5:DistributedBigDataSystemsandAnalytics
Demo1–4(Charleston1&2)
19:00-21:00 OrganizationEvent(InvitationOnly)
Day4:Thursday,June8,2017
Time Track1(SalonI)
Track2(SalonII)
Track3(Atlanta)
Track4(Columbia)
Track5(Nashville)
Track6(Charleston1)
7:30-8:30 ContinentalBreakfast(Foyer)8:30-9:30 Keynote3(PhoenixBallroom)9:30-10:00 CoffeeBreak(Foyer)
10:00-12:00Research14:
MobileandWirelessComputingSystemsII
Research15:SocialNetworksand
Crowdsourcing
Application9:DistributedSystemsandApplications
Application10:DistributedSystems
andServices
ShortPaper6:Security,Privacy,Trust,andFaultTolerancein
DistributedSystems
Tutorial2:SensorCloud:ACloudofSensor
Networks
12:00-13:30
BusinessLunchincludingAwardsandICDCS2018Announcements
(PhoenixBallroom)
13:30-15:30PlenaryPanel&
ConferenceClosingRemarks(PhoenixBallroom)
Tutorial2:SensorCloud:ACloudofSensor
Networks
![Page 21: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/21.jpg)
21
Day 1 – Monday, June 5, 2017
TenWorkshopsandtwotutorials(seeProgramataglanceandworkshopswebsiteandICDCS2017websitefordetails)
7:00-8:00 Monday, June 5, 2017 ContinentalBreakfastLocation:Foyer
8:00-9:30 Monday, June 5, 2017 Tutorial1:ServerlessProgramming(FunctionasaService)Location:Charleston1
PaulCastro(IBMT.J.WatsonResearchCenter),VatcheIshakian(BentleyUniversity),VinodMuthusamy(IBMT.J.WatsonResearchCenter),AleksanderSlominski(IBMT.J.WatsonResearchCenter)
Workshop:CCN-CPS,Session1Location:SalonII
SessionChair:NaderMohamed(MiddlewareTechnologiesLab.)
PoliciesGuidingCohesiveInteractionsamongInternetofThingswithCommunicationCloudandSocialNetworks HenryHexmoor(SouthernIllinoisUniversity)
EnhancedSecurityofBuildingAutomationSystemsThroughMicrokernel-BasedControllerPlatforms XiaolongWang(UniversityofSouthFlorida),RichardHabeeb(UniversityofSouthFlorida),XinmingOu(UniversityofSouthFlorida),SiddharthAmaravadi(KansasStateUniversity),JohnHatcliff(KansasStateUniversity),MasaakiMizuno(KansasStateUniversity),MitchellLNeilsen(KansasStateUniversity),RajRajagopalan(Honeywell),SrivatsanVaradarajan(HoneywellAerospaceAdvancedTechnologyLabs)
HighlevelDesignofaHomeAutonomousSystemBasedonCyberPhysicalSystemModeling BasmanAlhafidh(FloridaInstituteofTechnology),WilliamH.Allen(FloridaInstituteofTechnology)
Workshop:HotPOST,Session1Location:SalonVI
KeynoteSpeech:AMarkovGameTheoreticApproachforPowerGridSecurity CharlesA.Kamhoua(AirForceResearchLaboratory)
Router-basedBrokeringforSurrogateDiscoveryinEdgeComputing JulienGedeon(TechnischeUniversitätDarmstadt),ChristianMeurisch(TechnischeUniversitätDarmstadt),DishaBhat(TechnischeUniversitätDarmstadt),MichaelStein(TechnischeUniversitätDarmstadt),LinWang(TechnischeUniversitätDarmstadt),MaxMühlhäuser(TechnischeUniversitätDarmstadt)
ModelingtheSpreadofInfluenceforIndependentCascadeDiffusionProcessinSocialNetworks ZeshengChen(IndianaUniversity-PurdueUniversityFortWayne),KurtisTaylor(IndianaUniversity-PurdueUniversityFortWayne)
Workshop:PSBD,OpeningandInvitedTalksLocation:Atlanta
KeynoteSpeech:BigData-SecurityandPrivacy(andTransparency) ElisaBertino(PurdueUniversity)
InvitedTalk:SupportingTime-varyingPrivacywithSelf-emergingData BalajiPalanisamy(UniversityofPittsburgh)
Workshop:JCC,Session1Location:Columbia
HeterogeneousMalwareSpreadProcessinStarNetwork LiboJiao(TsinghuaUniversity),HaoYin(TsinghuaUniversity),DongchaoGuo(TsinghuaUniversity),YongqiangLyu(TsinghuaUniversity)
CostReductioninHybridCloudsforEnterpriseComputing
![Page 22: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/22.jpg)
22
BiyuZhou(InstituteofComputingTechnology,ChineseAcademyofSciences),FaZhang(InstituteofComputingTechnology,ChineseAcademyofSciences),JieWu(TempleUniversity),ZhiyongLiu(InstituteofComputingTechnology,ChineseAcademyofSciences)
DC-RSF:ADynamicandCustomizedReputationSystemFrameworkforJointCloudComputingFanghuaYe(SunYat-senUniversity),ZibinZheng(SunYat-senUniversity),ChuanChen(SunYat-senUniversity),YurenZhou(SunYat-senUniversity)
WebServiceApplianceBasedonUnikernel KaiYu(NationalLabforParallelandDistributedProcessing),ChengfeiZhang(NationalLabforParallelandDistributedProcessing),YunxiangZhao(NationalLabforParallelandDistributedProcessing)
AnalysisandEvaluationoftheGASModelforDistributedGraphComputation WangJinyan(NationalLabforParallelandDistributedProcessing),ZhangChengfei(NationalLabforParallelandDistributedProcessing)
TrafficSignsDetectionBasedonFasterR-CNN ZhongrongZuo(NationalLabforParallelandDistributedProcessing),KaiYu(NationalLabforParallelandDistributedProcessing),QiaoZhou(NationalLabforParallelandDistributedProcessing),XuWang(NationalLabforParallelandDistributedProcessing),TingLi(NationalLabforParallelandDistributedProcessing)
JCLedger:ABlockchainBasedDistributedLedgerforJointCloudComputing XiangFu(NationalUniversityofDefenseTechnology),HuaiminWang(NationalUniversityofDefenseTechnology),PeichangShi(NationalUniversityofDefenseTechnology),YingweiFu(NationalUniversityofDefenseTechnology),YijieWang(NationalUniversityofDefenseTechnology)
CorporationArchitectureforMultipleCloudServiceProvidersinJointCloudComputing PeichangShi(NationalUniversityofDefenseTechnology),HuaiminWang(NationalUniversityofDefenseTechnology),XikunYue(NationalUniversityofDefenseTechnology),ShilanYang(NationalUniversityofDefenseTechnology),ShangzhiYang(NationalUniversityofDefenseTechnology),YuxingPeng(NationalUniversityofDefenseTechnology)
SharingPrivacyDatainSemi-TrustworthyStoragethroughHierarchicalAccessControl YuzhaoWu(TsinghuaUniversity),YongqiangLyu(TsinghuaUniversity),QianFang(TsinghuaUniversity),GengZheng(TsinghuaUniversity),HaoYin(TsinghuaUniversity),YuanchunShi(TsinghuaUniversity)
Workshop:NSF-JSTLocation:Savannah
9:30-10:00 Monday, June 5, 2017 CoffeeBreakLocation:PhoenixBallroom
10:00-12:00 Monday, June 5, 2017 Tutorial1:ServerlessProgramming(FunctionasaService)Location:Charleston1
PaulCastro(IBMT.J.WatsonResearchCenter),VatcheIshakian(BentleyUniversity),VinodMuthusamy(IBMT.J.WatsonResearchCenter),AleksanderSlominski(IBMT.J.WatsonResearchCenter)
Workshop:CCN-CPS,Session2Location:SalonII
SessionChair:JameelaAlJaroodi(RobertMorrisUniversity)
ACyberPhysicalBuses-and-DronesMobileEdgeInfrastructureforLargeScaleDisasterEmergencyCommunications MamtaNarang(AucklandUniversityofTechnology),WilliamLiu(AucklandUniversityofTechnology),JairoAGutierrez(AucklandUniversityofTechnology),LucaChiaraviglio(UniversityofRomeTorVergata)
APerformanceComparisonofContainersandVirtualMachinesinWorkloadMigrationContext KumarGaurav(VMwareSoftwareIndiaPvtLtd),PavanKarkun(VMwareSoftwareIndiapvtLTD),Y.C.Tay(NationalUniversityofSingapore)
![Page 23: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/23.jpg)
23
TowardsService-OrientedMiddlewareforCyberPhysicalSystems NaderMohamed(MiddlewareTechnologiesLab.),SanjaLazarova-Molnar(UniversityofSouthernDenmark)
NetworkingandCommunicationinCyberPhysicalSystems ImadJawhar(UAEUniversity),JameelaAl-Jaroodi(RobertMorrisUniversity)
Workshop:ADSN,Session1:Keynote,Session2:AssuringTemporalFairnessandSecuringCommunicationLocation:SalonIV
KeynoteSpeech:DependabilityChallengesin5GCellularNetworks DouglasM.Blough(GeorgiaInstituteofTechnology)
UnderstandingandImprovingTemporalFairnessonanElectronicTradingVenue HaydenMelton(DeakinUniversity)
CertificateLessCryptography-basedRuleManagementProtocolforAdvancedMissionDeliveryNetworks JonghoWon(PurdueUniversity),AnkushSingla(PurdueUniversity),ElisaBertino(PurdueUniversity)
Workshop:HotPOST,Session2Location:SalonVI
ThankYouForBeingAFriend:AnAttackerViewonOnline-Social-Network-basedSybilDefenses DavidKoll(UniversityofGoettingen),MartinSchwarzmaier(UniversityofGoettingen),JunLi(UniversityofOregon),Xiang-YangLi(UniversityofScienceandTechnologyofChina),XiaomingFu(UniversityofGoettingen)
EfficientDynamicServiceFunctionChainCombinationofNetworkFunctionVirtualization WenkeYan(BeijingUniversityofPostsandTelecommunications),KonglinZhu(BeijingUniversityofPostsandTelecommunications),LinZhang(BeijingUniversityofPostsandTelecommunications),SixiSu(BeijingUniversityofPostsandTelecommunications)
WhenAugmentedRealitymeetsBigData CarlosBermejo(TheHongKongUniversityofScienceandTechnology),ZhanpengHuang(TheHongKongUniversityofScienceandTechnology),TristanBraud(TheHongKongUniversityofScienceandTechnology),PanHui(TheHongKongUniversityofScienceandTechnology)
SamplingBasedEfficientAlgorithmtoEstimatetheSpectralRadiusofLargeGraphs SamarAbbas(LahoreUniversityofManagementSciences),JuvariaTariq(LahoreUniversityofManagementSciences),ArifZaman(LahoreUniversityofManagementSciences),ImdadullahKhan(LahoreUniversityofManagementSciences)
ExtemporaneousMicro-MobileServiceExecutionWithoutCodeSharing ZhengSong(VirginiaTech),MinhLe(UtahStateUniversity),Young-WooKwon(UtahStateUniversity),EliTilevich(VirginiaTech)
PreventingColludingIdentityCloneAttacksinOnlineSocialNetworks GeorgesA.Kamhoua(FloridaInternationalUniversity),NikiPissinou(FloridaInternationalUniversity),S.S.Iyengar(FloridaInternationalUniversity),JonathanBeltran(FloridaInternationalUniversity),CharlesKamhoua(AirForceResearchLaboratory),BrandonLHernandez(UTRGV),LaurentNjilla(AirForceResearchLaboratory)
Workshop:PSBD,ResearchSessionLocation:Atlanta
Anovelgame-theoreticmodelforcontent-adaptiveimagesteganography QiLi(HunanUniversity),XinLiao(HunanUniversity),GuoyongChen(HunanUniversity),LipingDing(GuangzhouBranchofInstituteofSoftware,ChineseAcademyofScience)
AFine-grainedAccessControlSchemeforBigDataBasedonClassificationAttributes TengfeiYang(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),PeisongShen(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),XueTian(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),ChiChen(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences)
Social-AwareDecentralizationforEfficientandSecureMulti-PartyComputation YuzheTang(SyracuseUniversity),SuchetaSoundarajan(SyracuseUniversity)
StatisticalAnomalyDetectiononMetadataStreamsviaCommoditySoftwaretoProtectCompany ChristineChen(UniversityofPortland),JamesGurganus(MicroSystemsEngineering,Inc.)
Computationalimprovementsinparallelizedk-anonymousmicroaggregationoflargedatabases AhmadMohamadMezher(UniversitatPolitècnicadeCatalunya(UPC)),AlejandroGarcíaÁlvarez(UniversitatPolitècnicadeCatalunya(UPC)),DavidRebollo-Monedero(UniversitatPolitècnicadeCatalunya(UPC)),JordiForné(UniversitatPolitècnicadeCatalunya(UPC))
![Page 24: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/24.jpg)
24
Workshop:JCC,Session2Location:Columbia
AReliabilityBenchmarkforBigDataSystemsonJointCloudYingyingZheng(InstituteofSoftware,ChineseAcademyofSciences),LijieXu(InstituteofSoftware,ChineseAcademyofSciences),WeiWang(InstituteofSoftware,ChineseAcademyofSciences),WeiZhou(KSYUN),YingDing(ChangchunUniversityofScienceandTechnology)
UCPR:UserClassificationandInfluenceAnalysisinSocialNetwork CongZha(TsinghuaUniversity),YongqiangLv(TsinghuaUniversity)
AdaptiveRoutingAlgorithmforJointCloudVideoDelivery ZexunJiang(TsinghuaUniversity),HaoYin(TsinghuaUniversity)
TowardsEfficientResourceManagementinVirtualClouds BoAn(PekingUniversity),JunmingMa(PekingUniversity),DonggangCao(PekingUniversity),GangHuang(PekingUniversity)
MonitoringandBillingofALightweightCloudSystemBasedonLinuxContainer YujianZhu(PekingUniversity),JunmingMa(PekingUniversity),BoAn(PekingUniversity),DonggangCao(PekingUniversity)
Buildingemulationframeworkfornon-volatilememory GuoliangZhu(NationalUniversityofDefenseTechnology),KaiLu(NationalUniversityofDefenseTechnology),XiaopingWang(NationalUniversityofDefenseTechnology)
Seflow:EfficientFlowSchedulingforData-ParallelJobs QiaoZhou(NationalLabforParallelandDistributedProcessing),ZiyangLi(NationalLabforParallelandDistributedProcessing),PingZhong(CentralSouthUniversity),TianTian(NationalLabforParallelandDistributedProcessing),YuxingPeng(NationalLabforParallelandDistributedProcessing)
OnlineEncodingforErasure-CodedDistributedStorageSystems FangliangXu(NationalUniversityofDefenseTechnology),YijieWang(NationalUniversityofDefenseTechnology),XingkongMa(NationalUniversityofDefenseTechnology)
Workshop:NSF-JST,Session1Location:Savannah
AcceleratingBigDataInfrastructureandApplications KevinBrown(TokyoInstituteofTechnology),TianqiXu(TokyoInstituteofTechnology),KeitaIwabuchi(TokyoInstituteofTechnology),KentoSato(LawrenceLivermoreNationalLaboratory),AdamMoody(LawrenceLivermoreNationalLaboratory),KathrynMohror(LawrenceLivermoreNationalLaboratory),NikhilJain(LawrenceLivermoreNationalLaboratory),AbhinavBhatele(LawrenceLivermoreNationalLaboratory),MartinSchulz(LawrenceLivermoreNationalLaboratory),RogerPearce(LawrenceLivermoreNationalLaboratory),MayaGokhale(LawrenceLivermoreNationalLaboratory),SatoshiMatsuoka(TokyoInstituteofTechnology)
DisasterNetworkEvolutionUsingDynamicClusteringofTwitterData KrishnaKant(TempleUniversity),YilangWu(AizuUniversity),ShanshanZhang(TempleUniversity),JunboWang(AizuUniversity),AmitangshuPal(TempleUniversity)
Single-epochsupernovaclassificationwithdeepconvolutionalneuralnetworks AkisatoKimura(NTT),IchiroTakahashi(KavliIPMU,TheUniversityofTokyo),MasaomiTanaka(NationalAstronomicalObservatoryofJapan),NaokiYasuda(KavliIPMU,TheUniversityofTokyo),NaonoriUeda(NTT),NaokiYoshida(KavliIPMU,TheUniversityofTokyo)
EnablingLargeScaleDeliberationusingIdeationandNegotiation-SupportAgents KatsuhideFujita(TokyoUniversityofAgricultureandTechnology),TakayukiIto(NagoyaInstituteofTechnology),MarkKlein(MIT)
12:00-13:30 Monday, June 5, 2017 LunchLocation:Foyer
13:30-15:30 Monday, June 5, 2017 Workshop:CCN-CPS,Session3Location:SalonII
SessionChair:UttamGhosh(TennesseeStateUniversity)
OptimalDeploymentofChargingStationsforElectricVehicles:AFormalApproach
![Page 25: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/25.jpg)
25
AmarjitDatta(TennesseeTechnologicalUniversity),BrianLedbetter(TennesseeTechnologicalUniversity),MohammadAshiqurRahman(TennesseeTechnologicalUniversity)
FormalVerificationofControlStrategiesforaCyberPhysicalSystemAmjadGawanmeh(KhalifaUniversityofScienceandTechnology),AliAlwadi(AucklandUniversityofTechnology),SaziaParvin(UniversityofNewSouthWales)
LightweightDetectionandIsolationofBlackHoleAttacksinConnectedVehicles SamiAlbouq(OaklandUniversity),ErikFredericks(OaklandUniversity)
AnewthreatassessmentmethodforintegratinganIoTinfrastructureinaninformationsystem BrunoDorsemaine(OrangeLabs),Jean-PhilippeGaulier(OrangeLabs),Jean-PhilippeWary(OrangeLabs),NizarKheir(Thales),PascalUrien(TelecomParisTech)
Workshop:ADSN,Session3:NetworkAssuranceLocation:SalonIV
FaultySensorDataDetectioninWirelessSensorNetworksUsingLogisticalRegressionTianyuZhang(UniversityofHyogo),QianZhao(UniversityofHyogo),YukikazuNakamoto(UniversityofHyogo)
AnAdaptability-EnhancedRoutingMethodforMultipleGateway-basedWirelessSensorNetworksUsingSecureDispersedDataTransfer RyumaTani(HiroshimaCityUniversity),KentoAoi(HiroshimaCityUniversity),EitaroKohno(HiroshimaCityUniversity),YoshiakiKakuda(HiroshimaCityUniversity)
ProgressiveDownloadMethodBasedonTimer-DrivenRequestingSchemesUsingMultipleTCPFlowsonMultiplePaths HiroakiHoriba(HiroshimaCityUniversity),TokumasaHiraoka(HiroshimaCityUniversity),JunichiFunasaka(HiroshimaCityUniversity)
Workshop:PED-BGP,Session1Location:SalonVI
Keynotespeech:Application-awaredatadissemination BettinaKemme(McGillUniversity)
WED-SQL:ARelationalFrameworkforDesignandImplementationofProcess-AwareInformationSystems BrunoPadilha(UniversityofSaoPaulo),AndréLuisSchwerz(FederalUniversityofTechnology),RafaelLiberatoRoberto(FederalUniversityofTechnology)
QueryingWorkflowLogsYanTang(UniversityofCaliforniaatSantaBarbara),JianwenSu(UniversityofCaliforniaatSantaBarbara)
Ontheintegrationofevent-basedandtransaction-basedarchitecturesforSupplyChains ZhijieLi(IndianaUniversity–PurdueUniversityIndianapolis),HaoyanWu(IndianaUniversity–PurdueUniversityIndianapolis),BrianKing(IndianaUniversity–PurdueUniversityIndianapolis),ZinaBen-Miled(IndianaUniversity–PurdueUniversityIndianapolis),JohnWassick(TheDowChemicalCompany),JeffreyTazelaar(TheDowChemicalCompany)
Workshop:IoTCA,Session1Location:Atlanta
KeynoteSpeech:TheInternetofThings,People,andSystems:FromtheEdgetotheCloud SchahramDustdar(TUWien)
TowardsPrivacy-AwareSmartBuildings:Capturing,Communicating,andEnforcingPrivacyPoliciesandPreferences PrimalPappachan(UniversityofCaliforniaIrvine),MartinDegelingy(CarnegieMellonUniversity),RobertoYus(UniversityofCaliforniaIrvine),AnupamDasy(CarnegieMellonUniversity),SrutiBhagavatulay(CarnegieMellonUniversity),WilliamMelichery(CarnegieMellonUniversity),PardisEmamiNaeiniy(CarnegieMellonUniversity),ShikunZhangy(CarnegieMellonUniversity),LujoBauery(CarnegieMellonUniversity),AlfredKobsa(UniversityofCaliforniaIrvine),SharadMehrotra(UniversityofCaliforniaIrvine),NormanSadeh(CarnegieMellonUniversity),NaliniVenkatasubramanian(UniversityofCaliforniaIrvine)
DeployingData-DrivenSecuritySolutionsonResource-ConstrainedWearableIoTSystem HangCai(WorcesterPolytechnicInstitute),TianlongYun(WorcesterPolytechnicInstitute),JosiahHester(DartmouthCollege),KrishnaK.Venkatasubramanian(ClemsonUniversity)
AMotifbasedIoTFrameworkforDataEfficiency AkashSahoo(TexasA&MUniversity),RabiMahapatra(TexasA&MUniversity)
Workshop:WoSC,Session1
![Page 26: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/26.jpg)
26
Location:Columbia
KeynoteSpeech:ServerlessComputing:PatternsandRoadAhead RogerBarga(AmazonWebServices)
Ripple:HomeAutomationforResearchDataManagement RyanChard(ArgonneNationalLaboratory),KyleChard(ComputationInstitute,UniversityofChicagoandArgonneNationalLab),JasonAlt(NationalCenterforSupercomputingApplications),DilworthParkinson(LawrenceBerkeleyNationalLaboratory),SteveTuecke(ComputationInstitute,UniversityofChicagoandArgonneNationalLab),IanFoster(ArgonneNationalLaboratory&TheUniversityofChicago)
Pipsqueak:LeanLambdaswithLargeLibraries EdwardOakes(UniversityofWisconsin-Madison),LeonYang(UniversityofWisconsin-Madison),KevinHouck(UniversityofWisconsin-Madison),TylerHarter(MicrosoftGraySystemsLab),AndreaC.Arpaci-Dusseau(UniversityofWisconsin-Madison),RemziH.Arpaci-Dusseau(UniversityofWisconsin-Madison)
LeveragingtheServerlessArchitectureforSecuringLinuxContainersNiltonBila(IBM),PaoloDettori(IBM),AliKanso(IBM),YujiWatanabe(IBM),AlaaYoussef(IBM)
Workshop:NSF-JSTLocation:Savannah
15:30-16:00 Monday, June 5, 2017 CoffeeBreakLocation:Foyer
16:00-17:00 Monday, June 5, 2017 Workshop:CCN-CPS,Session4Location:SalonII
SessionChair:BrunoDorsemaine(OrangeLabs)
ASecurityFrameworkforSDN-enabledSmartPowerGrids UttamGhosh(TennesseeStateUniversity),PushpitaChatterjee(SRMRESEARCHINSTITUTE),SachinShetty(OldDominionUniversity)
Real-timeMonitoringSteamGeneratorsusingaHybridImagingSystem MahmoudMeribout(PetroleumInstitute),ImranSaied(PetroleumInstitute),EsraAlHosani(AdcoGroup)
SecuringbigDataEfficientlythroughMicroaggregationTechniqueandHuffmanCompression ShakilaMahjabinTonni(BangladeshArmyInternationalUniversityofScienceandTechnology),MohammadZahidurRahman(JahangirnagarUniversity),SaziaParvin(UniversityofNewSouthWales),AmjadGawanmeh(KhalifaUniversityofScienceandTechnology)
ModelBasedEnergyConsumptionAnalysisofWirelessCyberPhysicalSystems JingLiu(PekingUniversity),PingWang(PekingUniversity),JinlongLin(PekingUniversity),Chao-HsienChu(PennsylvanniaStateUniversity)
Workshop:ADSN,Session4:Panelon“AssuranceinInternetofThings(IoT)”Location:SalonIV
Moderator:EitaroKohno
Workshop:PED-BGP,Session2Location:SalonVI
CacheDOCS:ADynamicKey-ValueObjectCachingService JulienGascon-Samson(UniversityofBritishColumbia),MichaelCoppinger(McGillUniversity),FanJin(McGillUniversity),JörgKienzle(McGillUniversity),BettinaKemme(McGillUniversity)
WolfPath:Acceleratingiterativetraversing-basedgraphprocessingalgorithmsonGPU HuanzhouZhu(UniversityofWarwick),LigangHe(UniversityofWarwick)
ANovelAuction-basedQueryPricingSchema XingwangWang(JilinUniversity),XiaohuiWei(JilinUniversity),ShangGao(JilinUniversity),YuanyuanLiu(JilinUniversity),ZongpengLi(UniversityofCalgary)
BlockGraphChi:EnablingBlockUpdateinOut-of-coreGraphProcessing
![Page 27: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/27.jpg)
27
ZhiyuanShao(HuazhongUniversityofScienceandTechnology),ZhenjieMei(HuazhongUniversityofScienceandTechnology),XiaofengDing(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology)
Workshop:IoTCA,Session2Location:Atlanta
CoTWare:ACloudofThingsMiddleware JameelaAl-Jaroodi(RobertMorrisUniversity),NaderMohamed(MiddlewareTechnologiesLab.),ImadJawhar(MidcompResearchCenter)
SecuringtheInternetofThings:AMeta-StudyofChallenges,Approaches,andOpenProblemsMahmudHossain(UniversityofAlabamaatBirmingham),RagibHasan(UniversityofAlabamaatBirmingham),AnthonySkjellum(AuburnUniversity)
InternetofThingsFrameworkforSmartLearningAnalyticsAliYavari(SwinburneUniversityofTechnology),RezaSoltanpoor(RMITUniversity)
Workshop:WoSC,Session2Location:Columbia
ServerlessComputing:Design,Implementation,andPerformance GarrettMcGrath(UniversityofNotreDame),PaulR.Brenner(UniversityofNotreDame)
PaneldebateonthenoveltyandchallengesofserverlesscomputingParticipants:TBA
Workshop:NSF-JSTLocation:Savannah
17:00-18:00 Monday, June 5, 2017 Workshop:PED-BGP,Session2Location:SalonVI
IncrementalParallelComputingusingTransactionalModelinLarge-scaleDynamicGraphStructures AnandTripathi(UniversityofMinnesota),RahulR.Sharma(UniversityofMinnesota),ManuKhandelwal(UniversityofMinnesota),TanmayMehta(UniversityofMinnesota),VarunPandey(UniversityofMinnesota)
AgainstSigned-GraphDeanonymizationAttacks:PrivacyProtectionforSocialNetworks JianliangGao(CentralSouthUniversity),YuLiu(CentralSouthUniversity),PingZhong(CentralSouthUniversity),JianxinWang(CentralSouthUniversity)
![Page 28: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/28.jpg)
28
Day 2 – Tuesday, June 6, 2017
7:00-8:00 Tuesday, June 6, 2017 ContinentalBreakfastLocation:Foyer
8:00-8:30 Tuesday, June 6, 2017 ConferenceOpeningLocation:PhoenixBallroom
8:30-9:30 Tuesday, June 6, 2017 Keynote1byC.Mohan(IBMResearch)Location:PhoenixBallroom
SessionChair:MasaruKitsuregawa(UniversityofTokyo)
9:30-10:00 Tuesday, June 6, 2017 CoffeeBreakLocation:Foyer
10:00-12:00 Tuesday, June 6, 2017 Research1:DistributedFaultToleranceandDependabilityLocation:SalonI
SessionChair:MudhakarSrivatsa(IBMT.J.WatsonResearchCenter)
Timely,Reliable,andCost-EffectiveInternetTransportServiceusingDisseminationGraphs AmyBabay(JohnsHopkinsUniversity),EmilyWagner(JohnsHopkinsUniversity,LTNGlobalCommunications),MichaelDinitz(JohnsHopkinsUniversity),YairAmir(JohnsHopkinsUniversity,LTNGlobalCommunications)
Pronto:EfficientTestPacketGenerationforDynamicNetworkDataPlanes YuZhao(UniversityofKentucky),HuazheWang(UniversityofCaliforniaatSantaCruz),XinLin(UniversityofCaliforniaatSantaCruz),TingtingYu(UniverstiyofKentucky),ChenQian(UniversityofCaliforniaatSantaCruz)
Agar:ACachingSystemforErasure-CodedData RalucaHalalai(UniversityofNeuchâtel),PascalFelber(UniversityofNeuchâtel),Anne-MarieKermarrec(INRIA),FrançoisTaïani(IRISA)
Highperformancerecoveryforparallelstatemachinereplication OdoricoMendizabal(FURG),FernandoLuisDotti(PUCRS),FernandoPedone(UniversityofLugano)
OnDataParallelismofErasureCodinginDistributedStorageSystems JunLi(UniversityofToronto),BaochunLi(UniversityofToronto)
MeteorShower:MinimizingRequestLatencyforMajorityQuorum-basedDataConsistencyAlgorithmsinMultipleDataCenters YingLiu(KTHRoyalInstituteofTechnology),XiGuan(KTHRoyalInstituteofTechnology),VladimirVlassov(KTHRoyalInstituteofTechnology),SeifHaridi(KTHRoyalInstituteofTechnology)
Research2:DistributedOperatingSystemsandMiddlewareLocation:SalonII
SessionChair:PeterPietzuch(ImperialCollegeLondon)
LSbM-tree:Re-enablinghigh-speedcachinginDataManagementforMixedReadsandWrites DejunTeng(TheOhioStateUniversity),LeiGuo(Google),RubaoLee(TheOhioStateUniversity),FengChen(LouisianaStateUniversity),SiyuanMa(TheOhioStateUniversity),XiaodongZhang(TheOhioStateUniversity),YanfengZhang(NortheasternUniversity)
IncrementalTopologyTransformationforPublish/SubscribeSystemsUsingIntegerProgramming PooyaSalehi(TechnicalUniversityofMunich),KaiwenZhang(TechnicalUniversityofMunich),Hans-ArnoJacobsen(UniversityofToronto)
milliScope:aFine-GrainedMonitoringFrameworkforPerformanceDebuggingofn-TierWebServices
![Page 29: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/29.jpg)
29
Chien-AnLai(GeorgiaInstituteofTechnology),JoshKimball(GeorgiaInstituteofTechnology),TaoZhu(GeorgiaInstituteofTechnology),QingyangWang(LouisianaStateUniversity),CaltonPu(GeorgiaInstituteofTechnology)
Stark:OptimizingIn-MemoryComputingForDynamicDatasetCollections ShenLi(IBMResearch),MdTanvirAlAmin(UIUC),RaghuGanti(IBMResearch),MudhakarSrivatsa(IBMResearch),ShaohanHu(IBMResearch),YiranZhao(UIUC),TarekAbdelzaher(UIUC)
CRESON:CallableandReplicatedSharedObjectsoverNoSQL PierreSutra(TélécomSudParis,CNRS,UniversitéParis-Saclay,France),EtienneRivière(UniversityofNeuchatel),CristianCotes(UniversitatRoviraiVirgili),MarcSánchezArtigas(UniversitatRoviraiVirgili),PedroGarciaLopez(UniversitatRoviraiVirgili),EmmanuelBernard(RedHat),WilliamBurns(RedHat),GalderZamarreno(RedHat)
VirtualizedNetworkCodingFunctionsontheInternet LinquanZhang(UniversityofCalgary),ShangqiLai(TheUniversityofHongKong),ChuanWu(TheUniversityofHongKong),ZongpengLi(UniversityofCalgary),ChuanxiongGuo(MicrosoftResearch)
Vision1:InternetofThings,SmartCitiesandCyber-PhysicalSystemsLocation:Atlanta
SessionChair:LingLiu(GeorgiaInstituteofTechnology)
Observable-by-Design MasaruKitsuregawa(NationalInstituteofInformatics(NII)/InstituteofIndustrialScience,UniversityofTokyo)
AnArchitecturalVisionforaData-CentricIoT:RethinkingThings,TrustandClouds EveM.Schooler(Intel),DavidZage(Intel),JeffSedayao(Intel),HassnaaMoustafa(Intel),AndrewBrown(Intel),MorenoAmbrosin(UniversityofPadua)
EdgeComputingandIoTBasedResearchforBuildingSafeSmartCitiesResistanttoDisasters TeruoHigashino(OsakaUniversity),HirozumiYamaguchi(OsakaUniversity),AkihitoHiromori(OsakaUniversity),AkiraUchiyama(OsakaUniversity),KeiichiYasumoto(NaraInstituteofScienceandTechnology)
TheInternetofThingsandMultiagentSystems:DecentralizedIntelligenceinDistributedComputing MunindarSingh(NorthCarolinaStateUniversity),AmitChopra(LancasterUniversity)
InternetofThings:FromSmall-toLarge-ScaleOrchestration CharlesConsel(Inria/BordeauxINP),MilanKabac(ImperialCollege)
EdgeOS_H:AHomeOperatingSystemforInternetofEverything JieCao(WayneStateUniversity),LanyuXu(WayneStateUniversity),RaefAbdallah(WayneStateUniversity),WeisongShi(WayneStateUniversity)
Application1:Security,Privacy,TrustinDistributedSystemsLocation:Columbia
SessionChair:SongqingChen(GeorgeMasonUniversity)
PrivacyPreservingUser-basedRecommenderSystem ShahriarBadsha(RMITUniversity),XunYi(RMITUniversity),IbrahimKhalil(RMITUniversity),ElisaBertino(PurdueUniversity)
PrivacyPreservingOptimizationofParticipatorySensinginMobileCloudComputing YeYan(OaklandUniversity),DongHan(OaklandUniversity),TaoShu(AuburnUniversity)
SPHINX:APasswordStorethatPerfectlyHidesPasswordsfromItself MalihehShirvanian(UniversityofAlabamaatBirmingham),StanislawJarecki(UniversityofCaliforniaatIrvine),HugoKrawczyk(IBMResearch),NiteshSaxena(UniversityofAlabamaatBirmingham)
WhenSmartTVMeetsCRN:Privacy-preservingFine-grainedSpectrumAccess ChaowenGuan(StateUniversityofNewYorkatBuffalo),AzizMohaisen(StateUniversityofNewYorkatBuffalo),ZhiSun(StateUniversityofNewYorkatBuffalo),LuSu(StateUniversityofNewYorkatBuffalo),KuiRen(StateUniversityofNewYorkatBuffalo),YalingYang(VirginiaTech)
RevisitingSecurityRisksofAsymmetricScalarProductPreservingEncryptionandItsVariants WeipengLin(SimonFraserUniveristy),KeWang(SimonFraserUniversity),ZhilinZhang(SimonFraserUniversity),HongChen(RenminUniversityofChina)
AnAdversary-CentricBehaviorModelingofDDoSAttacks AnWang(GeorgeMasonUniversity),AzizMohaisen(SUNYBuffalo),SongqingChen(GeorgeMasonUniversity)
![Page 30: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/30.jpg)
30
Application2:SocialNetworksandCrowdsourcingLocation:Nashville
SessionChair:YaoLiu(SUNYBinghamton)
Anti-MaliciousCrowdsourcingUsingtheZero-DeterminantStrategy QinHu(BeijingNormalUniversity),ShenglingWang(BeijingNormalUniversity),LiranMa(TexasChristianUniversity),RongfangBie(BeijingNormalUniversity),XiuzhenCheng(GeorgeWashingtonUniversity)
JPR:ExploringJointPartitioningandReplicationforTrafficMinimizationinOnlineSocialNetworks JingyaZhou(SoochowUniversity),JianxiFan(SoochowUniversity)
OptimizingSourceSelectioninSocialSensinginthePresenceofInfluenceGraphs HuajieShao(UIUC),ShiguangWang(UIUC),ShenLi(UIUC),ShuochaoYao(UIUC),YiranZhao(UIUC),MdTanvirAlAmin(UIUC),TarekAbdelzaher(UIUC),LanceKaplan(UIUC)
DynamicContractDesignforHeterogenousWorkersinCrowdsourcingforQualityControl ChenxiQiu(PennsylvaniaStateUniversity),AnnaSquicciarini(PennsylvaniaStateUniversity),SarahRajtmajer(PennsylvaniaStateUniversity),JamesCaverlee(TexasA&MUniversity)
JointRequestBalancingandContentAggregationinCrowdsourcedCDN MingMa(TsinghuaUniversity),ZhiWang(TsinghuaUniversity),KunYi(TsinghuaUniversity),JiangchuanLiu(SouthChinaAgriculturalUniversity),LifengSun(TsinghuaUniversity)
Shrink:ABreastCancerRiskAssessmentModelBasedonMedicalSocialNetwork AliLi(UniversityofScienceandTechnologyBeijing),RuiWang(UniversityofScienceandTechnologyBeijing),LeiXu(UniversityofScienceandTechnologyBeijing)
Industry1:CloudDataCentersandPerformanceLocation:Charleston1
SessionChair:RaghuGanti(IBMT.J.WatsonResearchCenter)
Phoenix:Constraintawareschedulingforheterogeneousdatacenters PrashanthThinakaran(PennsylvaniaStateUniversity),JashwantRajGunasekaran(PennsylvaniaStateUniversity),BikashSharma(MicrosoftCorp),MahmutKandemir(PennsylvaniaStateUniversity),ChitaDas(PennsylvaniaStateUniversity)
DualScalingVMsandQueries:Cost-effectiveLatencyCurtailment JuanPérez(UniversityofMelbourne),RobertBirke(IBMResearchZurich),MathiasBjörkqvist(IBMResearchZurich),LydiaY.Chen(IBMResearchZurich)
Aframeworkforenablingsecurityservicescollaborationacrossmultipledomains DanielMigault(EricssonSecurityResearch),MarcosSimplicioJunior(EscolaPolitécnica),BrunoBarros(EscolaPolitécnica),MakanPourzandi(EricssonSecurityResearch),ThiagoAlmeida(EscolaPolitécnica),EwertonAndrade(EscolaPolitécnica),TerezaCarvalho(EscolaPolitécnica)
GroupClusteringUsingInter-GroupDissimilarities DebessayFesehayeKassa(VMware),LeninSingaravelu(Google),Chien-ChiaChen(VMware),XiaoboHuang(VMware),AmitabhaBanerjee(VMware),RuijinZhou(VMware),RajeshSomasundaran(VMware)
ComprehensiveMeasurementandAnalysisoftheUser-PerceivedI/OPerformanceinaProductionLeadership-ClassStorageSystemLipengWan(OakRidgeNationalLaboratory),MatthewWolf(OakRidgeNationalLaboratory),FeiyiWang(OakRidgeNationalLaboratory),JongYoulCho(OakRidgeNationalLaboratory),GeorgeOstruchov(OakRidgeNationalLaboratory),ScottKlasky(OakRidgeNationalLaboratory)
12:00-13:30 Tuesday, June 6, 2017 LunchLocation:PhoenixBallroom
13:30-15:30 Tuesday, June 6, 2017 Research3:SecurityandPrivacyinDistributedSystemsILocation:SalonI
SessionChair:RolandYap(NationalUniversityofSingapore)
![Page 31: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/31.jpg)
31
ConsensusRobustnessandTransactionDe-AnonymizationintheRippleCurrencyExchangeSystem AdrianoDiLuzio(SapienzaUniversityofRome),AlessandroMei(SapienzaUniversityofRome),JulindaStefa(SapienzaUniversityofRome)
LearningprivacyhabitsofPDSowners BikashSingh(UniversityofInsubria),BarbaraCarminati(universityofinsubria),ElenaFerrari(universityofinsubria)
City-Hunter:HuntingSmartphonesinUrbanAreas XuefengLiu(HongKongPolytechnicUniversity),JiaqiWen(HongKongPolytechnicUniversity),ShaojieTang(UniversityofTexasatDallas),JiannongCao(HongKongPolytechnicUniversity),JiaxingShen(HongKongPolytechnicUniversity)
WhenSeeingIsn'tBelieving:OnFeasibilityandDetectabilityofScapegoatinginNetworkTomography ShangqingZhao(UniversityofSouthFlorida),ZhuoLu(UniversityofSouthFlorida),CliffWang(NorthCarolinaStateUniversity/ArmyResearchOffice)
YouCanHearButYouCannotSteal:DefendingagainstVoiceImpersonationAttacksonSmartphones SiChen(UniversityatBuffalo/WestChesterUniversity),KuiRen(UniversityatBuffalo),SixuPiao(UniversityatBuffalo),CongWang(CityUniversityofHongKong),QianWang(WuhanUniversity),JianWeng(JinanUniversity),LuSu(UniveristyatBuffalo),AzizMohaisen(UniversityatBuffalo)
FlowReconnaissanceviaTimingAttacksonSDNSwitches ShengLiu(UniversityofNorthCarolinaatChapelHill),MichaelReiter(UniversityofNorthCarolinaatChapelHill),VyasSekar(CarnegieMellonUniversity)
Research4:CloudComputingandDataCenterSystemsLocation:SalonII
SessionChair:VladimirVlassov(KTHRoyalInstituteofTechnology)
AStudyofLong-TailLatencyinn-TierSystems:RPCvs.AsynchronousInvocations QingyangWang(LouisianaStateUniversity),Chien-AnLai(GeorgiaTech),YasuhikoKanemasa(FujitsuLaboratoriesLtd.),ShungengZhang(LouisianaStateUniversity),CaltonPu(GeorgiaTech)
RainorShine?-MakingSenseofCloudyReliabilityData IyswaryaNarayanan(ThePennsylvaniaStateUniversity),BikashSharma(Microsoft),DiWang(Microsoft),SriramGovindan(Microsoft),LauraCaulfield(Microsoft),AnandSivasubramaniam(ThePennsylvaniaStateUniversity),AmanKansal(Microsoft),JieLiu(Microsoft),BadriddineKhessib(Microsoft),KushagraVaid(Microsoft)
Right-sizingGeo-distributedDataCentersforAvailabilityandLatency IyswaryaNarayanan(ThePennsylvaniaStateUniversity),AmanKansal(Microsoft),AnandSivasubramaniam(ThePennsylvaniaStateUniversity)
PerformanceDrivenResourceSharingMarketsfortheSmallCloud Sung-HanLin(UniversityofSouthernCalifornia),RanjanPal(UniversityofSouthernCalifornia),MarcoPaolieri(UniversityofSouthernCalifornia),LeanaGolubchik(UniversityofSouthernCalifornia)
Fault-scalableVirtualizedInfrastructureManagement MukilKesavan(VMwareInc.),AdaGavrilovska(GeorgiaInstituteofTechnology),KarstenSchwan(GeorgiaInstituteofTechnology)
DeltaCFS:BoostingDeltaSyncforCloudStorageServicesbyLearningfromNFS QuanluZhang(PekingUniversity),ZhenhuaLi(TsinghuaUniversity),ZhiYang(PekingUniversity),ShenglongLi(PekingUniversity),YangzeGuo(PekingUniversity),YafeiDai(PekingUniversity),ShouyangLi(PekingUniversity)
Vision2:FutureNetworkingandCyberinfrastructureLocation:Atlanta
SessionChair:ManishParashar(RutgersUniversity)
AVisionforZero-HopNetworking(ZeN) MostafaAmmar(SchoolofComputerScience,GeorgiaTech),EllenZegura(SchoolofComputerScience,GeorgiaTech),YimengZhao(SchoolofComputerScience,GeorgiaTech)
StructuredOverlayNetworksforaNewGenerationofInternetServices AmyBabay(JohnsHopkinsUniversity),ClaudiuDanilov(BoeingResearchandTechnology),JohnLane(LTNGLobalCommunications),MichalMiskin-Amir(LTNGlobalCommunications,SpreadConceptsLLC),DanielObenshain(JohnsHopkinsUniversity),JohnSchultz(LTNGlobalCommunications,SpreadConceptsLLC),JonathanStanton(LTNGlobalCommunications,SpreadConceptsLLC),ThomasTantillo(JohnsHopkinsUniversity),YairAmir(JohnsHopkinsUniversity,LTNGlobalCommunications,SpreadConceptsLLC)
![Page 32: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/32.jpg)
32
EnsuringNetworkNeutralityforFutureDistributedSystems ThiagoGarrett(FederalUniversityofParana),SchahramDustdar(TUWien),LuisC.E.Bona(FederalUniversityofParana),EliasP.DuarteJr.(FederalUniversityofParana)
UncoveringtheUsefulStructuresofComplexNetworksinSocially-RichandDynamicEnvironments JieWu(TempleUniversity)
FutureNetworkingChallenges:TheCaseofMobileAugmentedReality TristanBraud(TheHongKongUniversityofScienceandTechnology),FarshidHassaniBijarbooneh(TheHongKongUniversityofScienceandTechnology),DimitrisChatzopoulos(TheHongKongUniversityofScienceandTechnology),PanHui(TheHongKongUniversityofScienceandTechnology)
SoftwareDefinedCyberinfrastructure IanFoster(ArgonneNationalLaboratoryandTheUniversityofChicago),BenBlaiszik(TheUniversityofChicago),KyleChard(ComputationInstitute,UniversityofChicagoandArgonneNationalLab),RyanChard(VictoriaUniversityofWellington)
Application3:InternetofThings,SmartCities,andCyber-PhysicalSystemsLocation:Columbia
SessionChair:GuanhuaYan(BinghamtonUniversity)
OpportunisticEnergySharingBetweenPowerGridandElectricVehicles:AGameTheory-BasedPricingPolicy AnkurSarker(UniversityofVirginia),ZhuozhaoLi(UniversityofVirginia),WilliamKolodzey(ClemsonUniversity),HaiyingShen(UniversityofVirginia)
EnergyEfficientObjectDetectioninCameraSensorNetworks TuanDao(UCRiverside),KarimKhalil(UCRiverside),AmitRoy-Chowdhury(UCRiverside),SrikanthKrishnamurthy(UCRiverside),LanceKaplan(U.S.ArmyResearchLaboratory)
DeepOpp:Context-awareMobileAccesstoSocialMediaContentonUndergroundMetroSystems DiWu(HunanUniversity&ImperialCollegeLondon),DmitriArkhipov(UniversityofCaliforniaIrvine),ThomasPrzepiorka(ImperialCollegeLondon),QiangLiu(DartmouthCollege),JulieMcCann(ImperialCollegeLondon),AmeliaRegan(UniversityofCaliforniaIrvine)
PhaseBeat:ExploitingCSIPhaseDataforVitalSignMonitoringwithCommodityWiFiDevices XuyuWang(AuburnUniversity),ChaoYang(AuburnUniversity),ShiwenMao(AuburnUniversity)
REX:RapidEnsembleClassificationSystemforLandslideDetectionusingSocialMediaAibekMusaev(UniversityofAlabama),DeWang(GeorgiaInstituteofTechnology),JiatengXie(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology)
TowardAnIntegratedApproachtoLocalizingFailuresinCommunityWaterNetworksQingHan(UCIrvine),PhuNguyen(UCIrvine),RonaldT.Eguchi(ImageCat),Kuo-LinHsu(UCIrvine),NaliniVenkatasubramanian(UCIrvine)
Application4:Mobile,Wireless,andEdgeComputingLocation:Nashville
SessionChair:ShiwenMao(AuburnUniversity)
MobiQoR:PushingtheEnvelopeofMobileEdgeComputingviaQuality-of-ResultOptimizationYongboLi(GeorgeWashingtonUniversity),YurongChen(GeorgeWashingtonUniversity),TianLan(GeorgeWashingtonUniversity),GuruVenkataramani(GeorgeWashingtonUniversity)
TruthfulAuctionsforUserDataAllowanceTradinginMobileNetworks ZhongxingMing(TsinghuaUniversity),MingweiXu(TsinghuaUniversity),NingWang(SurreyUniversity),BeijeGao(TsinghuaUniversity),QiLi(TsinghuaUniversity)
OnlineResourceAllocationforArbitraryUserMobilityinDistributedEdgeClouds LinWang(TUDarmstadt),LeiJiao(UniversityofOregon),JunLi(UniversityofOregon),MaxMühlhäuser(TUDarmstadt)
LeveragingTargetk-CoverageinWirelessRechargeableSensorNetworks PengzhanZhou(StonyBrookUniversity),CongWang(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity)
ReducingCellularSignalingTrafficforHeartbeatMessagesviaEnergy-EfficientD2DForwarding YanqiJin(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),XiaomengYi(HuazhongUniversityofScience&Technology),MinghuaChen(TheChineseUniversityofHongKong)
k-ProtectedRoutingProtocolinMulti-hopCognitiveRadioNetworks
![Page 33: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/33.jpg)
33
Chin-JungLiu(MichiganStateUniversity),LiXiao(MichiganStateUniversity)
ShortPaper1:DistributedOperatingSystems,Middleware,andAlgorithmsLocation:Charleston1
SessionChair:SripadNadagowda(IBMTJWatsonResearchCenter)
SRLB:ThePowerofChoicesinLoadBalancingwithSegmentRoutingYoannDesmouceaux(ÉcolePolytechnique),PierrePfister(CiscoSystems),JérômeTollet(CiscoSystems),MarkTownsley(CiscoSystems),ThomasClausen(ÉcolePolytechnique)
ImprovingEfficiencyofLinkClusteringonMulti-CoreMachines GuanhuaYan(BinghamtonUniversity)
S3:JointSchedulingandSourceSelectionforBackgroundTrafficinErasure-CodedStorage ShijingLi(GeorgeWashingtonUniversity),TianLan(GeorgeWashingtonUniversity),Moo-RyongRa(AT&TLabsResearch),RajeshPanta(AT&TLabsResearch)
OntheFeasibilityofInter-domainRoutingviaaSmallBrokerSet DongLin(HuaweiTechnologiesLtdCo.),DavidHui(HuaweiTechnologiesLtdCo.),WeijieWu(HuaweiTechnologiesLtdCo.),TingweiLiu(TheChineseUniversityofHongKong),YatingYang(BeijingInstituteofTechnology),YiWang(TsinghuaUniversity),JohnChi-ShingLui(ChineseUniversityofHongKong),GongZhang(HuaweiTechnologiesLtdCo.),YingtaoLi(HuaweiTechnologiesLtdCo.)
SubscriptionCoveringforRelevance-basedFilteringinContent-BasedPublish/SubscribeSystems KaiwenZhang(TechnischeUniversitatMunchen),VinodMuthusamy(IBMResearch),MohammadSadoghi(PurdueUniversity),Hans-ArnoJacobsen(UniversityofToronto)
WorkflowOptimizationinPAW MaximFilatov(UNIGE),VerenaKantere(UniversityofGeneva)
AFirstLookatInformationEntropy-BasedDataPricing XijunLi(ShanghaiJiaoTongUniversity),JianguoYao(ShanghaiJiaoTongUniversity),XueLiu(McGillUniversity),HaibingGuan(ShanghaiJiaoTongUniversity)
RestrospectiveLightweightDistributedSnapshotsUsingLooselySynchronizedClocks AlekseyCharapko(SUNYBuffalo),AilidaniAilijiang(SUNYBuffalo),MuratDemirbas(SUNYBuffalo),SandeepKulkarni(MichiganStateUniversity)
Power-AwarePopulationProtocols ChuanXu(LRI(CNRS/UPSud)),JannaBurman(LRI(CNRS/UPSud)),JoffroyBeauquier(LRI(CNRS/UPSud))
MultiPub:LatencyandCost-AwareGlobal-ScaleCloudPublish/Subscribe JulienGascon-Samson(McGillUniversity),JörgKienzle(McGillUniversity),BettinaKemme(McGillUniversity)
ReachabilityinBinaryMultithreadedProgramsIsPolynomial AlexanderMalkis(TechnischeUniversitätMünchen),SteffenBorgwardt(UCDavis)
AnEvent-LevelAbstractionforAchievingEfficiencyandFairnessinNetworkUpdate TingQu(NationalUniversityofDefenseTechnology),DekeGuo(NationalUniversityofDefenseTechnology),XiaominZhu(NationalUniversityofDefenseTechnology),JieWu(TempleUniversity),XiaoleiZhou(NationalUniversityofDefenseTechnology),ZhongLiu(NationalUniversityofDefenseTechnology)
15:30-16:00 Tuesday, June 6, 2017 CoffeeBreakLocation:Foyer
16:00-18:00 Tuesday, June 6, 2017 Research5:EdgeandFogComputingLocation:SalonI
SessionChair:WeisongShi(WayneStateUniversity)
Cachier:Edge-cachingforrecognitionapplications UtsavDrolia(CarnegieMellonUniversity),KatherineGuo(BellLabs),JiaqiTan(Nokia),RajeevGandhi(CarnegieMellonUniversity),PriyaNarasimhan(CarnegieMellonUniversity)
![Page 34: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/34.jpg)
34
ContentCentricPeerDataSharinginPervasiveEdgeComputingEnvironments XintongSong(PekingUniversity),YaodongHuang(StonyBrookUniversity),QianZhou(StonyBrookUniversity),FanYe(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity),XiaomingLi(PekingUniversity)
FLARE:CoordinatedRateAdaptationforHTTPAdaptiveStreaminginCellularNetworks YoungbinIm(UniversityofColoradoatBoulder),JinyoungHan(HanyangUniversity),JiHoonLee(JuniKorea),YoonKwon(Kakao),CarleeJoe-Wong(CarnegieMellonUniversity),TaekyoungKwon(SeoulNationalUniversity),SangtaeHa(UniversityofColoradoatBoulder)
NetworkedDroneCamerasforSportsStreaming XiaoliWang(PrincetonUniversity),AakankshaChowdhery(PrincetonUniversity),MungChiang(PrincetonUniversity)
Chronus:ConsistentDataPlaneUpdatesinTimedSDNs JiaqiZheng(NanjingUniversity),GuihaiChen(NanjingUniversity),StefanSchmid(AalborgUniversity),HaipengDai(NanjingUniversity),JieWu(TempleUniversity)
DistributedDeepNeuralNetworksovertheCloud,theEdgeandEndDevicesSuratTeerapittayanon(HarvardUniversity),BradleyMcDanel(HarvardUniversity),H.T.Kung(HarvardUniversity)
Research6:DistributedGreenComputingandEnergyManagementLocation:SalonII
SessionChair:GeraldFLofstead(SandiaNationalLab)
DynamicControlofFlowCompletionTimeForPowerEfficientDataCenterNetworks KuangyuZheng(TheOhioStateUniversity),XiaoruiWang(TheOhioStateUniversity)
OnEnergy-EfficientCongestionControlforMultipathTCP JiaZhao(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity),HaiyangWang(UniversityofMinnesotaDuluth)
AMechanismforCooperativeDemand-SideManagement GuangchaoYuan(Microsoft),Chung-WeiHang(IBM),MichaelHuhns(UniversityofSouthCarolina),MunindarSingh(NorthCarolinaStateUniversity)
AHierarchicalFrameworkofCloudResourceAllocationandPowerManagementUsingDeepReinforcementLearning NingLiu(SyracuseUniversity),ZheLi(SyracuseUniversity),ZhiyuanXu(SyracuseUniversity),JielongXu(SyracuseUniversity),ShengLin(SyracuseUniversity),QinruQiu(SyracuseUniversity),JianTang(SyracuseUniversity),YanzhiWang(SyracuseUniversity)
SunChase:Energy-EfficientRoutePlanningforSolar-PoweredEVs. LanduJiang(McGillUniversity),YuHua(HuazhongUniversityofScienceandTechnology),ChenMa(McGillUniversity),XueLiu(McGillUniversity)
Vision3:NextGenerationCloudandEdgeServicesLocation:Atlanta
SessionChair:ManfredHauswirth(TUBerlin)
ComputingintheContinuum:CombiningPervasiveDevicesandServicestoSupportData-drivenApplications ManishParashar(RutgersUniversity),MoustafaAbdelbaky(RutgersUniversity),MengsongZou(RutgersUniversity),AliRezaZamani(RutgersUniversity),EduardRenart(RutgersUniversity),JavierDiaz-Montes(RutgersUniversity)
Decision-drivenExecution:ADistributedResourceManagementParadigmfortheAgeofIoT TarekAbdelzaher(UIUC),TanvirAlAmin(UIUC),AmotzBar-Noy(UIUC),WilliamDron(BBN),RameshGovindan(USC),ReginaldHobbs(ARL),ShaohanHu(IBM),Jung-EunKim(UIUC),ShuochaoYao(UIUC),YiranZhao(UIUC)
ACTiCLOUD:EnablingtheNextGenerationofCloudApplications GeorgiosGoumas(NationalTechnicalUniversityofAthens),KonstantinosNikas(ComputingSystemsLaboratory,NTUA),EwnetuBayuhLakew(Dept.ofComputingScience,UmeaUniversity),ChristosKotselidis(TheUniversityofManchester),VasileiosKarakostas(ComputingSystemsLaboratory,NTUA),AtleVesterkjaer(Numascale),EinarRustad(Numascale),JohnGoodacre(Kaleao),AndrewAttwood(Kaleao),MichailFlouris(OnApp),JohnThomson(OnApp),NikosFoutris(TheUniversityofManchester),MikelLujan(TheUniversityofManchester),YingZhang(MonetDBSolutions),PanagiotisKoutsourakis(MonetDBSolutions),MartinKersten(MonetDBSolutions),JimWebber(NeoTechnology),DavideGrohmann(NeoTechnology),ErikElmroth(Dept.ofComputingScience,UmeaUniversity),LuisTomas(Dept.ofComputingScience,UmeaUniversity),NectariosKoziris(NationalTechnicalUniversityofAthens)
JointCloud:ACross-CloudCooperationArchitectureforIntegratedInternetServiceCustomization HuaiminWang(NationalUniversityofDefenseTechnology),PeichangShi(NationalUniversityofDefenseTechnology),YimingZhang(NationalUniversityofDefenseTechnology)
![Page 35: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/35.jpg)
35
SupportingDataAnalyticsApplicationsWhichUtilizeCognitiveServices ArunIyengar(IBMResearch)
TrillionOperationsKey-ValueStorageEngine:RevisitingtheMissionCriticalAnalyticsStorageSoftwareStack SangeethaSeshadri(IBMAlmadenResearchCenter),LawrenceChiu(IBMAlmadenResearchCenter),PaulMuench(IBMAlmadenResearchCenter)
ShortPaper2:CloudandDataCenterSystemsandNetworksLocation:Columbia
SessionChair:QingyangWang(LouisianaStateUniversity)
DCM:DynamicConcurrencyManagementforScalingn-TierApplicationsinCloud HuiChen(LouisianaStateUniversity),QingyangWang(LouisianaStateuniversity),BalajiPalanisamy(UniversityofPittsburgh),PengchengXiong(Hortonworks)
MorePeak,LessDifferentiation:TowardsAPricing-awareOnlineControlFrameworkforInter-DatacenterTransfers WenxinLi(DalianUniversityofTechnology),XiaoboZhou(TiajjinUniversity),KeqiuLi(DalianUniversityofTechnology),HengQi(DalianUniversityofTechnology),DekeGuo(NationalUniversityofDefenceTechnology)
RobustMulti-TenantServerConsolidationintheCloudforDataAnalyticsWorkloads JosephMate(UniversityofWaterloo),KhuzaimaDaudjee(UniversityofWaterloo),ShahinKamali(MITCSAIL)
Flow-AwareAdaptivePacingtoMitigateTCPIncastinDatacenterNetworks ShaojunZou(CentralSouthUniversity),JiaweiHuang(CentralSouthUniversity),YutaoZhou(CentralSouthUniversity),JianxinWang(CentralSouthUniversity),TianHe(UniversityofMinnesota)
Real-TimePowerCyclinginVideoonDemandDataCentresusingOnlineBayesianPrediction VicentSanzMarco(LancasterUniversity),ZhengWang(LancasterUniversity),BarryPorter(LancasterUniversity)
ADistributedAccessControlSystemforCloudFederations ShorouqAlansari(UniversityofSouthampton),FedericaPaci(UniversityofSouthampton),VladimiroSassone(UniversityofSouthampton)
Voyager:CompleteContainerStateMigration ShripadNadgowda(IBMTJWatsonResearchCenter),SahilSuneja(IBMTJWatsonResearchCenter),NiltonBila(IBMTJWatsonResearchCenter),CanturkIsci(IBMTJWatsonResearchCenter)
Keddah:CapturingHadoopNetworkBehaviour JieDeng(QueenMaryUniversityLondon),GarethTyson(QueenMary),FélixCuadrado(QueenMaryUniversityofLondon),SteveUhlig(QueenMaryUniversityofLondon)
AScalableandDistributedApproachforNFVServiceChainCostMinimization ZijunZhang(UniversityofCalgary),ZongpengLi(UniversityofCalgary),ChuanWu(UniversityofHongKong),ChuanheHuang(WuhanUniversity)
ElasticPaxos:ADynamicAtomicMulticastProtocol SamuelBenz(UniversitàdellaSvizzeraitaliana),FernandoPedone(UniversitàdellaSvizzeraitaliana)
BoostingTheBenefitsOfHybridSDN WenWang(McGillUniversity),WenboHe(McMasterUniversity),JinshuSu(NationalUniversityofDefenseTechnology)
AdoptingSDNSwitchBuffer:BenefitsAnalysisandMechanismDesign FuliangLi(NortheasternUniversity),JiannongCao(TheHongKongPolytechnicUniversity),XingweiWang(NortheasternUniversity),YinchuSun(NortheasternUniversity),TianPan(BeijingUniversityofPostsandTelecommunication),XuefengLiu(TheHongKongPolytechnicUniversity)
ShortPaper3:InternetofThings,SmartCities,andCyber-PhysicalSystemsLocation:Nashville
SessionChair:NanLi(CSIRO)
IOTSENTINEL:AutomatedDevice-TypeIdentificationforSecurityEnforcementinIoT MarkusMiettinen(TechnischeUniversitatDarmstadt),SamuelMarchal(AaltoUniversity),IbbadHafeez(UniversityofHelsinki),N.Asokan(AaltoUniversity),Ahmad-RezaSadeghi(TechnischeUniversitatDarmstadt),SasuTarkoma(UniversityofHelsinki)
EfficientZ-orderEncodingBasedMulti-modelDataCompressioninWSNs XiaofeiCao(MissouriUniversityofScienceandTechnology),SanjayMadria(MissouriUniversityofScienceandTechnology),TakahiroHara(OsakaUniversity)
PTrack:EnhancingtheApplicabilityofPedestrianTrackingwithWearables
![Page 36: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/36.jpg)
36
YonghangJiang(CityUniversityofHongKong),ZhenjiangLi(CityUniversityofHongKong),JianpingWang(CityUniversityofHongKong)
SourceLocationPrivacy-AwareDataAggregationSchedulingforWirelessSensorNetworks JackKirton(UniversityofWarwick),MatthewBradbury(TheUniversityofWarwick),ArshadJhumka(UniversityofWarwick)
VelocityOptimizationofPureElectricVehicleswithTrafficDynamicsConsideration LiuwangKang(UniversityofVirginia),HaiyingShen(UniversityofVirginia),AnkurSarker(UniversityofVirginia)
PIANO:Proximity-basedUserAuthenticationonVoice-PoweredInternet-of-ThingsDevices NeilZhenqiangGong(IowaStateUniversity),AltayOzen(IowaStateUniversity),YuWu(UCDavis),XiaoyuCao(IowaStateUniversity),RichardShin(UCBerkeley),DawnSong(UCBerkeley),HongxiaJin(SamsungResearchAmerica),XuanBao(GoogleInc.)
CategoryInformationCollectioninRFIDSystems JiaLiu(NanjingUniversity),ShigangChen(UniversityofFlorida),BinXiao(TheHongKongPolytechnicUniversity),YanyanWang(NanjingUniversity),LijunChen(NanjingUniversity)
ScalableRole-basedDataDisclosureControlfortheInternetofThings AliYavari(RMITUniversity),ArezouSoltaniPanah(RMITUniversity),DimitriosGeorgakopoulos(SwinburneUniversityofTechnology),PremPrakashJayaraman(SwinburneUniversityofTechnology),RonvanSchyndel(RMITUniversity)
Multi-representationbasedDataProcessingArchitectureforIoTApplications VaibhavArora(UniversityofCalifornia,SantaBarbara),FaisalNawab(UniversityofCalifornia,SantaBarbara),DivyakantAgrawal(UniversityofCalifornia,SantaBarbara),AmrElAbbadi(UniversityofCalifornia,SantaBarbara)
LongTermSensingviaBatteryHealthAdaptation GregJackson(ImperialCollegeLondon),ZhijinQin(ImperialCollegeLondon),JulieAMcCann(ImperialCollegeLondon)
DetectingTimeSynchronizationAttacksinCyber-PhysicalSystemswithMachineLearningTechniques JingxuanWang(TheUniversityofHongKong),WentingTu(ShanghaiUniversityofFinanceandEconomics),LucasC.K.Hui(TheUniversityofHongKong),SiuMingYiu(TheUniversityofHongKong),EricKeWang(HarbinInstituteofTechnologyShenzhenGraduateSchool)
Speed-basedLocationTrackinginUsage-basedAutomotiveInsurance LuZhou(ShanghaiJiaoTongUniversity),QingrongChen(ShanghaiJiaoTongUniversity),ZutianLuo(ShanghaiJiaoTongUniversity),HaojinZhu(ShanghaiJiaoTongUniversity),CailianChen(ShanghaiJiaoTongUniversity)
Poster1-4Location:Charleston1and2(Lighteningtalksfrom16:00to17:00)
SessionChairs:GeraldFLofstead(SandiaNationalLab)andBalajiPalanisamy(UniversityofPittsburg)
Location:PhoenixBallroom(Displayfrom17:00to19:00)
Poster1:DistributedApplicationsCluster
TowardVehicleSensing:Anintegratedapplicationwithsparsevideocamerasandintelligenttaxicabs YangWang(UniversityofScienceandTechnologyofChina),WujiChen(UniversityofScienceandTechnologyofChina),WeiZheng(Sanofi-AventisUSLLC),HeHuang(SoochowUniversity),WenZhang(UniversityofScienceandTechnologyofChina),HengchangLiu(UniversityofScienceandTechnologyofChina)
SegmentationofTimeSeriesbasedonKineticCharacteristicsforStorageConsumptionPrediction BeibeiMiao(Baidu,Inc),YuChen(Baidu,Inc),XueboJin(SchoolofComputerandInformationEngineering,BeijingTechnologyandBusinessUniversity),BoWang(Baidu,Inc),XianpingQu(Baidu,Inc),DongWang(Baidu,Inc),ShiminTao(Baidu,Inc),ZhiZang(Baidu,Inc)
AMulti-stageHierarchicalWindowModelwithApplicationtoReal-TimeGraphAnalysis SachiniJayasekara(UniversityofMelbourne),ShanikaKarunasekera(UniversityofMelbourne),AaronHarwood(UniversityofMelbourne)
DynamicPricingatElectricVehicleChargingStationsforQueueingDelayReduction XiaoshanSun(UniversityofScienceandTechnologyofChina),PengXu(UniversityofScienceandTechnologyofChina),JinyangLi(UniversityofScienceandTechnologyofChina),HengchangLiu(UniversityofScienceandTechnologyofChina),WeiZheng(Sanofi-Aventis)
PairwiseRankingAggregationbyNon-interactiveCrowdsourcingwithBudgetConstraints ChangjiangCai(StevensInstituteofTechnology),HaipeiSun(StevensInstituteofTechnology),BoxiangDong(MontclairStateUniversity),BoZhang(StevensInstituteofTechnology),TingWang(LehighUniversity),WendyHuiWang(StevensInstituteofTechnology)
Buffer-BasedReinforcementLearningforAdaptiveStreaming YueZhang(SUNYBinghamton),YaoLiu(SUNYBinghamton)
Thecaseforusingcontent-centricnetworkingfordistributinghigh-energyphysicssoftware
![Page 37: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/37.jpg)
37
MohammadAlhowaidi(UniversityofNebraska-Lincoln),ByravRamamurthy(UniversityofNebraska-Lincoln),BrianBockelman(UniversityofNebraska-Lincoln),DavidSwanson(UniversityofNebraska-Lincoln)
LAVEA:Latency-awareVideoAnalyticsonEdgeComputingPlatformShanheYi(CollegeofWilliamandMary),ZijiangHao(CollegeofWilliamandMary),QingyangZhang(WayneStateUniversity),QuanZhang(WayneStateUniversity),WeisongShi(WayneStateUniversity),QunLi(CollegeofWilliamandMary)
CompleteToleranceRelationbasedFillingAlgorithmusingSpark JinglingYuan(WuhanUniversityofTechnology),YaoXiang(WuhanUniversityofTechnology),XianZhong(WuhanUniversityofTechnology),MinchengChen(WuhanUniversityofTechnology),TaoLi(UniversityofFlorida)
Poster2:SecurityandPrivacyCluster
TowardsSecurePublicDirectoryforPrivacy-PreservingDataSharingAminFallahi(SyracuseUniversity),XiLiu(SyracuseUniversity),YuzheTang(SyracuseUniversity),ShuangWang(UCSD),RuiZhang(ChineseAcademyofSciences)
AnonymousRoutingtoMaximizeDeliveryRatesinDTNs KazuyaSakai(TokyoMetropolitanUniversity),Min-TeSun(NationalCentralUniversity),Wei-ShinnKu(AuburnUniversity),JieWu(TempleUniversity)
EvaluatingConnectionResiliencefortheOverlayNetworkKademliaHennerHeck(UniversitätKassel),OlgaKieselmann(UniversitätKassel),ArnoWacker(UniversitätKassel)
Shortfall-basedOptimalSecurityProvisioningforInternetofThings AntoninoRullo(UniversityofCalabria),EdoardoSerra(BoiseStateUniversity),JorgeLobo(UniversitatPompeaFabra),ElisaBertino(PurdueUniversity)
GroupDifferentialPrivacy-preservingDisclosureofMulti-levelAssociationGraphs BalajiPalanisamy(UniversityofPittsburgh),ChaoLi(UniversityofPittsburgh),PrashantKrishnamurthy(UniversityofPittsburgh)
TrackingInformationFlowinCyber-PhysicalSystems StefanGries(UniversityofDuisburg-Essen),MarcHesenius(UniversityofDuisburg-Essen),VolkerGruhn(UniversityofDuisburg-Essen)
Privacy-preservingMatchmakinginGeosocialNetworkswithUntrustedServers QiuxiangDong(ArizonaStateUniversity),DijiangHuang(ArizonaStateUniversity)
You’veBeenTricked!AUserStudyoftheEffectivenessofTyposquattingTechniques JeffreySpaulding(SUNYBuffalo),ShambhuUpadhyaya(SUNYBuffalo),AzizMohaisen(SUNYBuffalo)
Real-timeDetectionofIllegalFileTransfersintheCloud AdamBowers(MissouriUniversityofScienceandTechnology),DanLin(MissouriUniversityofScienceandTechnology),AnnaSquicciarini(ThePennsylvaniaStateUniversity),AliHurson(MissouriUniversityofScienceandTechnology
EyesoftheSwarm:Streamers'DetectioninBitTorrent DanielSilva(FluminenseFederalUniversity),AntonioRocha(FluminenseFederalUniversity)
Poster3:CloudsandVirtualizationCluster
Loadpredictionforenergy-awareschedulingforCloudscomputingplatformsAlexandreDambreville(LRI),JoannaTomasik(CentraleSupélec),JohanneCohen(LRI-CNRS),FabienDufoulon(LRI)
Learn-as-you-gowithMegh:EfficientLiveMigrationofVirtualMachines DebabrotaBasu(NationalUniversityofSingapore),XiayangWang(InstituteofParallelandDistributedSystems,ShanghaiJiaoTongUniversity),YangHong(ShanghaiJiaoTongUniversity),HaiboChen(ShanghaiJiaoTongUniversity),StephaneBressan(NationalUniversityofSingapore)
Machine-LearningBasedPerformanceEstimationforDistributedParallelApplicationsinVirtualizedHeterogeneousClusters SeontaeKim(UNIST),NguyenPham(UNIST),WoongkiBaek(UNIST),Young-RiChoi(UNIST)
IncrementalelasticityforNoSQLdatastores AntonisPapaioannou(ICS-FORTHandUniversityofCrete),KostasMagoutis(ICS-FORTHandUniversityofIoannina)
AFrameworkforEfficientEnergySchedulingofSparkWorkloads StathisMaroulis(AthensUniversityofEconomicsandBusiness),NikosZacheilas(AthensUniversityofEconomicsandBusiness),VanaKalogeraki(AthensUniversityofEconomicsandBusiness)
TowardsaCompleteVirtualDataCenterEmbeddingAlgorithmusingHybridStrategy MPGilesh(NationalInstituteofTechnologyCalicut),SDMadhuKumar(NationalInstituteofTechnologyCalicut),LillykuttyJacob(NationalInstituteofTechnologyCalicut),UmeshBellur(IndianInstituteofTechnologyBombay)
![Page 38: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/38.jpg)
38
FederatingConsistencyforPartition-ProneNetworks BenjaminBengfort(UniversityofMaryland),PeteKeleher(UniversityofMaryland)
Mitigatingnesting-agnostichypervisorpoliciesinderivativeclouds ChandraPrakash(IITBombay),Prashanth(IITBombay),PurushottamKulkarni(IITBombay),UmeshBellur(IITBombay)
ANovelArchitectureforEfficientFogtoCloudDataManagementinSmartCities AmirSinaeepourfard(UPC),JordiGarcia(UPC),XavierMasip-Bruin(UPC),EvaMarin-Tordera(UPC)
Networklet:ConceptandDeploymentShengZhang(NanjingUniversity),YuLiang(NanjingUniversityofPostsandTelecommunications),ZhuzhongQian(NanjingUniversity),MingjunXiao(UniversityofScienceandTechnologyofChina),JieWu(TempleUniversity),FanyuKong(AntFinancial),SangluLu(NanjingUniversity)
OptimisticCausalConsistencyforGeo-ReplicatedKey-ValueStores KristinaSpirovska(EPFL),DiegoDidona(EPFL),WillyZwaenepoel(EPFL)
AutomatedPerformanceEvaluationforMulti-TierCloudServiceSystemsSubjecttoMixedWorkloads XudongZhao(ShandongUniversity),LizhenCui(ShandongUniversity),JiweiHuang(BeijingUniversityofPostsandTelecommunications),ShijunLiu(ShandongUniversity),LeiLiu(ShandongUniversity),CaltonPu(GeorgiaTech)
DecentralisedRuntimeMonitoringforAccessControlSystemsinCloudFederations MdSadekFerdous(UniversityofSouthampton),AndreaMargheri(UniversityofSouthampton),FedericaPaci(UniversityofSouthampton),MuYang,VladimiroSassone(UniversityofSouthampton)
Poster4:DistributedSystemsandNetworkingCluster
DuoFS:AnAttemptatEnergy-SavingandRetainingReliabilityofStorageSystems ShuYin(HunanUniversity)
AProposalofanEfficientTrafficMatrixEstimationunderPacketDrops KoheiWatabe(NagaokaUniversityofTechnology),ToruMano(NTTNetworkInnovationLaboratories),KimihiroMizutani(NTTNetworkInnovationLaboratories),OsamuAkashi(NTTNetworkInnovationLaboratories),KenjiNakagawa(NagaokaUniversityofTechnology),TakeruInoue(NTTNetworkInnovationLaboratories)
StragglerMitigationforDistributedBehavioralSimulation EmanBinKhunayn(UniversityofMelbourne),ShanikaKarunasekera(UniversityofMelbourne),HairuoXie(UniversityofMelbourne),KotagiriRamamohanarao(UniversityofMelbourne)
SupportingResourceControlforActorSystemsinAkka AhmedAbdelMoamen(UniversityofSaskatchewan),DezhongWang(UniversityofSaskatchewan),NadeemJamali(UniversityofSaskatchewan)
ADistributedOperatingSystemNetworkStackandDeviceDriverforMulticores BMSaifAnsary(ECE,VirginiaTech),AntonioBarbalace(ECE,VirginiaTech),BinoyRavindran(ECE,VirginiaTech),ThomasLazor(ECE,VirginiaTech),Ho-RenChuang(ECE,VirginiaTech)
CachePotentialityofMONs:APrime PeiyanYuan(HenanNormalUniversity),HonghaiWu(HenanUniversityofScienceandTechnology),XiaoyanZhao(HenanNormalUniversity),ZhengnanDong(HenanNormalUniversity)
Oak:User-TargetedWebPerformance MarcelFlores(NorthwesternUniversity),AlexanderWenzel(NorthwesternUniversity),AleksandarKuzmanovic(NorthwesternUniversity)
Ctrl-A:ASelf-*DistributedandIn-bandSDNControlPlaneMarcoCanini(UniversitécatholiquedeLouvain),IosifSalem(ChalmersUniversityofTechnology),LironSchiff(TelAvivUniversity),EladMichaelSchiller(ChalmersUniversityofTechnology),StefanSchmid(AalborgUniversity&TUBerlin)
18:00-20:00 Tuesday, June 6, 2017 ReceptionLocation:PhoenixBallroom
![Page 39: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/39.jpg)
39
Day 3 – Wednesday, June 7, 2017
7:30-8:30 Wednesday, June 7, 2017 ContinentalBreakfastLocation:Foyer
8:30-9:30 Wednesday, June 7, 2017 Keynote2byM.TamerÖzsu(UniversityofWaterloo)Location:PhoenixBallroom
SessionChair:KarlAberer(EPFL)
9:30-10:00 Wednesday, June 7, 2017 CoffeeBreakLocation:Foyer
10:00-12:00 Wednesday, June 7, 2017 Research7:InternetofThings,SmartCities,andCyber-PhysicalSystemsLocation:SalonI
SessionChair:JoaoEduardoFerreira(UniversityofSaoPaulo)
PersistentTrafficMeasurementThroughVehicle-to-InfrastructureCommunications HeHuang(SoochowUniversity),Yu-ESun(SoochowUniversity),ShigangChen(UniversityofFlorida),HongliXu(UniversityofScienceandTechnologyofChina),YianZhou(Google)
TagBreathe:MonitorBreathingwithCommodityRFIDSystems YuxiaoHou(TheHongKongPolytechnicUniversity),YanwenWang(TheHongKongPolytechnicUniversity),YuanqingZheng(TheHongKongPolytechnicUniversity)
Double-EdgedSword:IncentivizedVerifiableProductPathQueryforRFID-enabledSupplyChain SaiyuQi(XidianUniversity),YuanqingZheng(TheHongKongPolytechnicUniversity),XiaofengChen(XidianUniversity),JianfengMa(XidianUniversity),YongQi(XianJiaotongUniversity)
TowardsAccurateCorruptionEstimationinZigBeeUnderCross-TechnologyInterference GongLongChen(ZhejiangUniversity),WeiDong(ZhejiangUniversity),ZhiweiZhao(UniversityofElectronicScienceandTechnologyofChina),TaoGu(RMITUniversity)
UnseenActivityRecognition:AHierarchicalActiveTransferLearningApproach MohammadArifUlAlam(UniversityofMarylandBaltimoreCounty),NirmalyaRoy(UniversityofMarylandBaltimoreCounty)
RFIPad:EnablingCost-efficientandDevice-freeIn-airHandwritingusingPassiveTags HanDing(Xi'anJiaotongUniversity),ChenQian(UniversityofCaliforniaSantaCruz),JinsongHan(Xi'anJiaotongUniversity),GeWang(Xi'anJiaotongUniversity),WeiXi(Xi'anJiaotongUniversity),KunZhao(Xi'anJiaotongUniversity),JizhongZhao(Xi'anJiaotongUniversity)
Research8:MobileandWirelessComputingSystemsILocation:SalonII
SessionChair:KarthikSundaresan(NECLaboratoriesAmerica)
RobustIncentiveTreeDesignforMobileCrowdsensing XiangZhang(ArizonaStateUniversity),GuoliangXue(ArizonaStateUniversity),RuozhouYu(ArizonaStateUniversity),DejunYang(ColoradoSchoolofMines),JianTang(SyracuseUniversity)
WearLock:UnlockingYourPhoneviaAcousticsusingSmartwatch ShanheYi(CollegeofWilliamandMary),ZhengruiQin(NorthwestMissouriStateUniversity),NancyCarter(CollegeofWilliamandMary),QunLi(CollegeofWilliamandMary)
ModelingMobileCodeAccelerationintheCloud
![Page 40: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/40.jpg)
40
HuberFlores(UniversityofOulu),XiangSu(UniversityofOulu),VassilisKostakos(UniversityofOulu),JukkaRiekki(UniversityofOulu),EemilLagerspetz(UniversityofHelsinki),SasuTarkoma(HelsinkiUniversityofTechnology),PanHui(HKUST),YongLi(TsinghuaUniversity),JukkaManner(AaltoUniversity)
E-Android:ANewEnergyProfilingToolforSmartphones XingGao(CollegeofWIlliamandMary),DachuanLiu(CollegeofWIlliamandMary),DaipingLiu(UniversityofDelaware),HainingWang(UniversityofDelaware),AngelosStavrou(GeorgeMasonUniversity)
LocalandLow-CostWhiteSpaceDetection AhmedSaeed(GeorgiaInstituteofTechnology),KhaledHarras(CarnegieMellonUniversity),EllenZegura(GeorgiaInstituteofTechnology),MostafaAmmar(GeorgiaInstituteofTechnology)
GeneralAnalysisofIncentiveMechanismsforPeer-to-PeerTransmissions:AQuantumGamePerspective WeimanSun(BeijingNormalUniversity),ShenglingWang(BeijingNormalUniversity)
Vision4:SecurityandTrustinFutureSystemsLocation:Atlanta
SessionChair:LingLiu(GeorgiaInstituteofTechnology)
HowComputerScienceRiskstoLoseItsInnocence,andShouldAttempttoTakeResponsibilityKarlAberer(EPFL)
ACognitivePolicyFrameworkforNext-GenerationDistributedFederatedSystems-ConceptsandResearchDirections ElisaBertino(PurdueUniversity),SeraphinCalo(IBM),MarounTouma(IBM),DineshVerma(IBM),ChristopherWilliams(UKDSTL),BrianRivera(ArmyResearchLabs)
MachinetoMachineTrustinSmartCities MargaretLoper(GeorgiaTechResearchInstitute),BrianSwenson(GeorgiaInstituteofTechnology)
LateralThinkingforTrustworthyApps HermannHärtig(TechnischeUniversitätDresden),MichaelRoitzsch(TechnischeUniversitätDresden),CarstenWeinhold(TechnischeUniversitätDresden),AdamLackorzynski(TechnischeUniversitätDresden)
RumorInitiatorDetectioninInfectedSignedNetworks JiaweiZhang(UniversityofIllinoisatChicago),CharuC.Aggarwal(IBMT.J.WatsonResearchCenter),PhilipS.Yu(UniversityofIllinoisatChicago)
AddressingSmartphone-basedMulti-factorAuthenticationviaHardware-rootedTechnologiesZhongjieBa(TheStateUniversityofNewYorkatBuffalo),KuiRen(TheStateUniversityofNewYorkatBuffalo)
Application5:CloudComputingandDataCenterSystemsLocation:Columbia
SessionChair:YuanyuanYang(StonyBrookUniversity)
Multi-ResourceLoadBalancingforVirtualNetworkFunctions TaoWang(HuazhongUniversityofScience&Technology),HongXu(CityUniversityofHongKong),FangmingLiu(HuazhongUniversityofScienceandTechnology)
Learningfromfailureacrossmultipleclusters:Atrace-drivenapproachtounderstanding,predicting,andmitigatingjobterminations NosaybaEl-Sayed(MIT),HongyuZhu(UniversityofToronto),BiancaSchroeder(UniversityofToronto)
RBAY:AScalableandExtensibleInformationPlaneforFederatingDistributedDatacenterResources XinChen(GeorgiaInstituteofTechnology),LitingHu(FloridaInternationalUniversity),DouglasM.Blough(GeorgiaInstituteofTechnology),MichaelA.Kozuch(IntelLabsPittsburgh),MatthewWolf(OakRidgeNationalLaboratory)
Task-awareTCPinDataCenterNetworks SenLiu(CentralSouthUniversity),JiaweiHuang(CentralSouthUniversity),YutaoZhou(CentralSouthUniversity),JianxinWang(CentralSouthUniversity),TianHe(UniversityofMinnesota)
LimitationsofLoadBalancingMechanismsforN-TierSystemsinthePresenceofMillibottlenecks TaoZhu(GeorgiaInstituteofTechnology),JackLi(GeorgiaInstituteofTechnology),JoshKimball(GeorgiaInstituteofTechnology),JunheePark(IndianaUniversity),Chien-AnLai(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology),QingyangWang(LouisianaStateUniversity)
PerformanceAnalysisofCloudComputingCentersServingParallelizableRenderingJobsUsingM/M/c/rQueuingSystems
![Page 41: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/41.jpg)
41
XiulinLi(ShandongUniversity),LiPan(ShandongUniversity),JiweiHuang(BeijingUniversityofPostsandTelecommunications),ShijunLiu(ShandongUniversity),YuliangShi(ShandongUniversity),CaltonPu(GeorgiaInstituteofTechnology)
Application6:BigDataSystemsandDistributedDataManagementandAnalyticsLocation:Nashville
SessionChair:ZhengruiQin(NorthwestMissouriStateUniversity)
EvaluationofDeepLearningFrameworksoverDifferentHPCArchitectures ShayanShams(LouisianaStateUniversity),RichardPlatania(LouisianaStateuniversity),KisungLee(LouisianaStateUniversity),Seung-JongPark(LouisianaStateUniversity)
OnAchievingEfficientDataTransferforGraphProcessinginGeo-DistributedDatacenters AmelieChiZhou(InriaRennes),ShadiIbrahim(InriaRennes),BingshengHe(NationalUniversityofSingapore)
GBooster:TowardsAccelerationofGPU-intensiveMobileApplications ElliottWen(VictoriaUniversityofWellington),BryanNg(VictoriaUniversityofWellington),WinstonSeah(VictoriaUniversityofWellington),XueLiu(McGillUniversity),JiannongCao(TheHongKongPolytechnicUniversity),XuefengLiu(HuangzhongUniversityofScienceandTechnology)
Scalingk-NearestNeighborsQueries(Therightway) AtoshumSamuelCahsai(UniversityOfGlasgow),NikosNtarmos(UniversityOfGlasgow),ChristosAnagnostopoulos(UniversityofGlasgow),PeterTriantafillou(UniversityOfGlasgow)
ParallelizingBigDeBruijnGraphConstructiononHeterogeneousProcessors ShuangQiu(TheHongKongUniversityofScienceandTechnology),QiongLuo(TheHongKongUniversityofScienceandTechnology)
Private,yetPractical,MultipartyDeepLearning XinyangZhang(LehighUniversity),ShoulingJi(ZhejiangUniversity),HuiWang(StevensInstituteofTechnology),TingWang(LehighUniversity)
Industry2:MobileComputingandInternetofThingsLocation:Charleston1
SessionChair:RobertBirke(IBMResearchZurich)
OntheLimitsofSubsamplingofLocationTraces MudhakarSrivatsa(IBMT.J.WatsonResearchCenter),RaghuGanti(IBMT.J.WatsonResearchCenter),PrasantMohapatra(UCDavis)
SOM-TC:Self-organizingmapforHierarchicalTrajectoryClustering PranitaDewan(IBMTJWatsonResearchCenter),RaghuGanti(IBMTJWatsonResearchCenter),MudhakarSrivatsa(IBMTJWatsonResearchCenter)
ProcessingEncryptedandCompressedTime-SeriesData MatúšHarvan(EnovosLuxembourgS.A.),SamuelKimoto(OpenSystems),ThomasLocher(ABBCorporateResearch),Yvonne-AnnePignolet(ABBCorporateResearch),JohannesSchneider(UniversityofLiechtenstein)
CalvinConstrained-AFrameworkforIoTApplicationsinHeterogeneousEnvironments AmardeepMehta(UmeåUniversity),RamiBaddour(UniversitádellaSvizzeraitaliana),FredrikSvensson(EricssonResearch),HaraldGustafsson(EricssonResearch),ErikElmroth(UmeåUniversity)
12:00-13:30 Wednesday, June 7, 2017 ConferenceLuncheonLocation:PhoenixBallroom
13:30-15:30 Wednesday, June 7, 2017 Research9:DistributedBigDataSystemsLocation:SalonI
SessionChair:KisungLee(LouisianaStateUniversity)
High-PerformanceandResilientKey-ValueStorewithOnlineErasureCodingforBigDataWorkloads DiptiShankar(TheOhioStateUniversity),XiaoyiLu(TheOhioStateUniversity),DhabaleswarPanda(TheOhioStateUniversity)
ModelingandAnalyzingLatencyintheMemcachedsystemWenxueCheng(TsinghuaUniversity),FengyuanRen(TsinghuaUniversity),WanchunJiang(CentralSouthUniversity),TongZhang(TsinghuaUniversity)
![Page 42: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/42.jpg)
42
SpeculativeSlotReservation:EnforcingServiceIsolationforDependentData-ParallelComputations ChenChen(HKUST),WeiWang(HKUST),BoLi(HKUST)
OptimizingShuffleinWide-AreaDataAnalytics ShuhaoLiu(UniversityofToronto),HaoWang(UniversityofToronto),BaochunLi(UniversityofToronto)
JobSchedulingwithoutPriorInformationinBigDataProcessingSystems ZhimingHu(UniversityofToronto),BaochunLi(UniversityofToronto),ZhengQin(InstituteofHighPerformanceComputing),RickSiowMongGoh(InstituteofHighPerformanceComputing)
DistributedLoadBalancinginKey-ValueNetworkedCachesSikderHuq(TheUniversityofIowa),ZubairShafiq(TheUniversityofIowa),SukumarGhosh(TheUniversityofIowa),AmirKhakpour(VerizonDigitalMediaServices),HarkeeratBedi(VerizonDigitalMediaServices)
Research10:DistributedAlgorithmsandTheoryILocation:SalonII
SessionChair:HaiyingShen(UniversityofVirginia)
CognitiveContext-awareDistributedStorageOptimizationinMobileCloudComputing:AStableMatchingbasedApproach DongHan(OaklandUniversity),YeYan(OaklandUniversity),TaoShu(AuburnUniversity),LiuqingYang(ColoradoStateUniversity),ShuguangCui(UniversityofCalifornia,Davis)
FairCachingAlgorithmsforPeerDataSharinginPervasiveEdgeComputingEnvironments YaodongHuang(StonyBrookUniversity),XintongSong(PekingUniversity),FanYe(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity),XiaomingLi(PekingUniversity)
Latency-DrivenCooperativeTaskComputinginMulti-UserFog-RadioAccessNetworks Ai-ChunPang(NationalTaiwanUniversity),Wei-HoChung(AcademiaSinica),Te-ChuanChiu(NationalTaiwanUniversity),JunshanZhang(ArizonaStateUniversity)
ApproximationandOnlineAlgorithmsforNFV-EnabledMulticastinginSDNs ZichuanXu(UniversityCollegeLondon),WeifaLiang(TheAustralianNationalUniversity),MeitianHuang(TheAustralianNationalUniversity),MikeJia(TheAustralianNationalUniversity),SongGuo(TheHongKongPolytechnicUniversity),AlexGalis(UniversityCollegeLondon)
DistributedAuctionsforTaskAssignmentandSchedulinginMobileCrowdsensingSystems ZhuojunDuan(GeorgiaStateUniversity),WeiLi(GeorgiaStateUniversity),ZhipengCai(GeorgiaStateUniversity)
EffectiveMobileDataTradinginSecondaryAd-hocMarketwithHeterogeneousandDynamicEnvironment HengkySusanto(HuaweiFutureNetworkTheoryLab),HonggangZhang(UniversityofMassachusettsBoston),ShingYipHo(ShareMedia),BenyuanLiu(UniversityofMassachusettsLowell)
Vision5:FutureDistributedSystemsLocation:Atlanta
SessionChair:ManishParashar(RutgersUniversity)
EnablingwideareadataanalyticswithCollaborativeDistributedProcessingPipes(CDPPs) AnjaFeldmann(TUBerlin),ManfredHauswirth(TUBerlin),VolkerMarkl(TUBerlin)
TheMillibottleneckTheoryofPerformanceBugs,andItsExperimentalVerification CaltonPu(GeorgiaInstituteofTechnology),JoshuaKimball(GeorgiaInstituteofTechnology),Chien-AnLai(GeorgiaInstituteofTechnology),TaoZhu(GeorgiaInstituteofTechnology),JackLi,JunheePark,QingyangWang,DeepalJayasinghe,PengchengXiong,SimonMalkowski,QinyiWu,GueyoungJung,YounggyunKoh,GalenSwint
Exacution:EnhancingScientificDataManagementforExascale ScottKlasky(OakRidgeNationalLaboratory),EricSuchyta(OakRidgeNationalLab),MarkAinsworth(BrownUniversity),QingLiu(NewJerseyInstituteofTechnology),BenWhitney(BrownUniversity),MatthewWolf(OakRidgeNationalLaboratory),JongChoi(OakRidgeNationalLaboratory),IanFoster(ArgonneNationalLaboratory),MarkKim(OakRidgeNationalLaboratory),JeremyLogan(UniversityOfTennesseeKnoxville),KshitijMehta(OakRidgeNationalLaboratory),ToddMunson(ArgonneNationalLaboratory),GeorgeOstrouchov(OakRidgeNationalLaboratory),ManishParashar(RutgersUniversity),NorbertPodhorszk(OakRidgeNationalLaboratory),DavidPugmire(OakRidgeNationalLaboratory),LipengWan(OakRidgeNationalLaboratory)
HardwareAccelerationLandscapeforDistributedReal-timeAnalytics:VirtuesandLimitations MohammadrezaNajafi(TechnischeUniversitatMunchen),KaiwenZhang(TechnischeUniversitatMunchen),Hans-ArnoJacobsen(UniversityofToronto),MohammadSadoghi(PurdueUniversity)
![Page 43: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/43.jpg)
43
CoordinatingDistributedSpeakingObjects MarcoLippi(DISMI–UniversitàdiModenaeReggioEmilia),MarcoMamei(DISMI–UniversitàdiModenaeReggioEmilia),StefanoMariani(DISMI–UniversitàdiModenaeReggioEmilia),FrancoZambonelli(DISMI–UniversitàdiModenaeReggioEmilia)
Model-DrivenDomain-SpecificMiddleware FabioCosta(FederalUniversityofGoias),KarlMorris(TempleUniversity),FabioKon(UniversityofSãoPaulo),PeterClarke(FloridaInternationalUniversity)
Application7:DistributedMiddlewareSystemsLocation:Columbia
SessionChair:QingyangWang(LouisianaStateUniversity)
FastandFlexibleNetworkingforMessage-orientedMiddleware LarsKroll(KTHRoyalInstituteofTechnology),AlexandruA.Ormenisan(KTHRoyalInstituteofTechnology),JimDowling(KTHRoyalInstituteofTechnology)
TailCut:PowerReductionunderQualityandLatencyConstraintsinDistributedSearchSystems Chih-HsunChou(UniversityofCalifornia,Riverside),LaxmiBhuyan(UniversityofCalifornia,Riverside),ShaoleiRen(UniversityofCalifornia,Riverside)
StoArranger:EnablingEfficientUsageofCloudStorageServicesonMobileDevices YongshuBai(SUNYBinghamton),YifanZhang(SUNYBinghamton)
CharacterizingPerformanceandEnergy-efficiencyofTheRAMCloudStorageSystem YacineTaleb(Inria),ShadiIbrahim(Inria),GabrielAntoniu(Inria),ToniCortes(BarcelonaSupercomputingCenter)
ProactivelySecureCloud-EnabledStorageKarimEldefrawy(HughesResearchLab),TylerKaczmarek(UniversityofCalifornia,Irvine),SkyFaber(UniversityofCalifornia,Irvine)
BEES:Bandwidth-andEnergy-EfficientImageSharingforReal-timeSituationAwareness PengfeiZuo(HuazhongUniversityofScienceandTechnology),YuHua(HuazhongUniversityofScienceandTechnology),XueLiu(McGillUniverisity),DanFeng(HuazhongUniversityofScienceandTechnology),WenXia(HuazhongUniversityofScienceandTechnology),ShundeCao(HuazhongUniversityofScienceandTechnology),JieWu(HuazhongUniversityofScienceandTechnology),YuanyuanSun(HuazhongUniversityofScienceandTechnology),YunchengGuo(HuazhongUniversityofScienceandTechnology)
Application8:DistributedSystemsandOptimizationsLocation:Nashville
SessionChair:AzizMohaisen(StateUniversityofNewYorkatBuffalo)
TransparentFault-ToleranceusingIntra-MachineFull-Software-StackReplication GiulianoLosa(VirginiaTech),AntonioBarbalace(VirginiaTech),YuzhongWen(VirginiaTech),MarinaSadini(VirginiaTech),Ho-RenChuang(VirginiaTech),BinoyRavindran(VirginiaTech)
Apreventiveauto-parallelizationapproachforelasticstreamprocessing RolandKottoKombi(UniversityClaudeBernard),NicolasLumineau(UniversitédeLyon),PhilippeLamarre(INSALyon)
DependableCloudResourceswithGuardian BaraAbusalah(PurdueUniversity),DerekSchatzlein(PurdueUniversity),JulianJamesStephen(PurdueUniversity),MasoudSaeidaArdekani(PurdueUniversity),PatrickEugster(PurdueUniversity)
ACommunication-awareContainerRe-distributionApproachforHighPerformanceVNFs YuchaoZhang(TsinghuaUniversity),YusenLi(NankaiUniversity),KeXu(TsinghuaUniversity),DanWang(HongKongPolytechnicUniversity),MinghuiLi(Baidu),XuanCao(Baidu),QingqingLiang(Baidu)
MinimizingCostinIaaSCloudsviaScheduledInstanceReservation QiushiWang(NanyangTechnologicalUniversity),MingMingTan(NanyangTechnologicalUniversity),XueyanTang(NanyangTechnologicalUniversity),WentongCai(NanyangTechnologicalUniversity)
EfficientDistributedCoordinationatWAN-scale AilidaniAilijiang(SUNYBuffalo),AlekseyCharapko(SUNYBuffalo),MuratDemirbas(SUNYBuffalo),BekirOguzTurkkan(SUNYBuffalo),TevfikKosar(SUNYBuffalo)
ShortPaper4:Mobile,Wireless,Edge,andCrowdComputingLocation:Charleston1
![Page 44: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/44.jpg)
44
SessionChair:LakshmishRamaswamy(UniversityofGeorgia)
Onefficientoffloadingcontrolincloudradioaccessnetworkwithmobileedgecomputing TongLi(TsinghuaUniversity),ChathuraSarathchandraMagurawalage(UniversityofEssex),KezhiWang(UniversityofEssex),KeXu(TsinghuaUniversity),KunYang(UniversityofEssex),HaiyangWang(UniversityofMinnesotaatDuluth)
LocationPrivacyinMobileEdgeClouds TingHe(PennsylvaniaStateUniversity),ErtugrulCiftcioglu(ArmyResearchLaboratory),ShiqiangWang(IBM),KevinChan(ArmyResearchLaboratory)
ApproximationDesignforCooperativeRelayDeploymentinWirelessNetworks HaotianWang(ShanghaiJiaoTongUniversity),ShileiTian(ShanghaiJiaoTongUniversity),XiaofengGao(ShanghaiJiaoTongUniversity),LidongWu(UniversityofTexasatTyler),GuihaiChen(ShanghaiJiaoTongUniversity)
DispersingSocialContentinMobileCrowdthroughOpportunisticContacts LeiZhang(SimonFraserUniversity),FengWang(TheUniversityofMississippi),JiangchuanLiu(SouthChinaAgriculturalUniversity)
ALightweightRecommendationFrameworkforMobileUser'sLinkSelectioninDenseNetwork JiWang(NationalUniversityofDefenseTechnology),XiaominZhu(NationalUniversityofDefenseTechnology),WeidongBao(NationalUniversityofDefenseTechnology),GuanlinWu(NationalUniversityofDefenseTechnology)
MakingSmartphoneSmartonDemandforLongerBatteryLife MarcoBrocanelli(TheOhioStateUniversity),XiaoruiWang(TheOhioStateUniversity)
FADEWICH:FastDeauthenticationovertheWirelessChannel GiulioLovisotto(UniversityofOxford),MauroConti(UniversityofPadua),IvanMartinovic(UniversityofOxford),GeneTsudik(UniversityofCalifornia,Irvine)
CognitiveWirelessCharger:Sensing-BasedReal-TimeFrequencyControlForNear-FieldWirelessCharging Sang-YoonChang(UniversityofColoradoColoradoSprings),SristiLakshmiSravanaKumar(AdvancedDigitalSciencesCenter),Yih-ChunHu(UniversityofIllinoisatUrbana-Champaign)
DensityandMobility-drivenEvaluationofBroadcastAlgorithmsforMANETs RazielCarvajalGómez(UniversityofNeuchatel),IntiGonzalez-Herrera(LaBRI/UniversityofBordeaux),Yérom-DavidBromberg(UniversityofRennes1),LaurentRéveillère(UniversityofBordeaux),EtienneRivière(UniversityofNeuchatel)
Energy-AwareCPUFrequencyScalingforMobileVideoStreaming WenjieHu(ThePennsylvaniaStateUniversity),GuohongCao(ThePennsylvaniaStateUniversity)
CrazyCrowdSourcingtoMitigateResourceScarcity NovaAhmed(NorthSouthUniversity),MdMahfuzurRahmanSiddiquee(IndependentResearcher),RefayaKarim(NorthSouthUniversity),MohsinaZaman(IndependentResearcher),SayedMahmudulAlam(NorthSouthUniversity),SyedFahimAsraf(NorthSouthUniversity)
DetectingRogueAPwiththeCrowdWisdom TongqingZhou(NationalUniversityofDefenseTechnology),ZhipingCai(NationalUniversityofDefenseTechnology),BinXiao(TheHongKongPolytechnicUniversity),YueyueChen(NationalUniversityofDefenseTechnology),MingXu(NationalUniversityofDefenseTechnology)
15:30-16:00 Wednesday, June 7, 2017 CoffeeBreakLocation:Foyer
16:00-18:00 Wednesday, June 7, 2017 Research11:SecurityandPrivacyinDistributedSystemsIILocation:SalonI
SessionChair:KuiRen(StateUniversityofNewYorkatBuffalo)
Kalis-ASystemforKnowledge-drivenAdaptableIntrusionDetectionfortheInternetofThings DanieleMidi(PurdueUniversity),AntoninoRullo(UniversityofCalabria),AnandMudgerikar(PurdueUniversity),ElisaBertino(PurdueUniversity)
FuzzyExtractorsforBiometricIdentification NanLi(CSIRO),FuchunGuo(UniversityofWollongong),YiMu(UniversityofWollongong),WillySusilo(UniversityofWollongong),SuryaNepal(CSIRO)
SmartphonePrivacyLeakageofSocialRelationshipsandDemographicsfromSurroundingAccessPoints
![Page 45: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/45.jpg)
45
ChenWang(StevensInstituteofTechnology),ChuyuWang(StevensInstituteofTechnology),YingyingChen(StevensInstituteofTechnology),LeiXie(NanjingUniversity),SangluLu(NanjingUniversity)
EV-Matching:BridgingLargeVisualDataandElectronicDataforEfficientSurveillanceGangLi(TheOhioStateUniversity),FanYang(TheOhioStateUniversity),GuoxingChen(TheOhioStateUniversity),QiangZhai(TheOhioStateUniversity),XinfengLi(TheOhioStateUniversity),JinTeng(TheOhioStateUniversity),JundaZhu(UniversityofMacau),DongXuan(TheOhioStateUniversity),BiaoChen(UniversityofMacau),WeiZhao(UniversityofMacau)
AdaptiveReconnaissanceAttackswithNear-OptimalParallelBatching XiangLi(UniversityofFlorida),JohnathanSmith(UniversityofFlorida),MyThai(UniversityofFlorida)
AchievingStrongPrivacyinOnlineSurvey YouZhou(UniversityofFlorida),YianZhou(UniversityofFlorida),ShigangChen(UniversityofFlorida),SamuelS.Wu(UniversityofFlorida)
Research12:CloudComputingandDistributedDataAnalyticsLocation:SalonII
SessionChair:KekeChen(WrightStateUniversity)
ServiceOverlayForestEmbeddingforSoftware-DefinedCloudNetworks Jian-JhihKuo(AcademiaSinica),Shan-HsiangShen(NationalTaiwanUniversityofScienceandTechnology),Ming-HongYang(UniversityofMinnesota),De-NianYang(AcademiaSinica),Ming-JerTsai(NationalTsingHuaUniversity),Wen-TsuenChen(AcademiaSinica)
JointOptimizationofChainPlacementandRequestSchedulingforNetworkFunctionVirtualization QixiaZhang(HuazhongUniversityofScience&Technology),YikaiXiao(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),JohnChiShingLui(ChineseUniversityofHongKong),JianGuo(HuazhongUniversityofScience&Technology),TaoWang(HuazhongUniversityofScience&Technology)
BIGCacheAbstractionforCacheNetworks EmanRamadan(UniversityofMinnesota),ArvindNarayanan(UniversityofMinnesota),Zhi-LiZhang(UniversityofMinnesota),RunhuiLi(HuaweiFutureNetworkTheoryLab),GongZhang(HuaweiFutureNetworkTheoryLab)
DistributedQRdecompositionframeworkfortrainingSupportVectorMachines JyotikrishnaDass(TexasA&MUniversity),V.N.S.PrithviSakuru(TexasA&MUniversity),VivekSarin(TexasA&MUniversity),RabiN.Mahapatra(TexasA&MUniversity)
DistributivelyComputingRandomWalkBetweennessCentralityinLinearTime Qiang-ShengHua(HuazhongUniversityofScienceandTechnology),MingAi(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),DongxiaoYu(HuazhongUniversityofScienceandTechnology),XuanhuaShi(HuazhongUniversityofScienceandTechnology)
DeGPar:LargeScaleTopicDetectionusingNode-CutPartitioningonDenseWeightedGraphs KambizGhoorchian(RoyalInstituteofTechnology(KTH)),SarunasGirdzijauskas(RoyalInstituteofTechnology(KTH)),FatemehRahimian(SwedishInstituteofComputerScience(SICS))
Vision6:InnovationinBigDataSystemsLocation:Atlanta
SessionChair:ManfredHauswirth(TUBerlin)
OntheDesignofaBlockchainPlatformforClinicalTrialandPrecisionMedicine ZonyinShae(ASIAUniversity,Taiwan),JeffreyTsai(ASIAUniversity,Taiwan)
TowardsDataflow-basedGraphAccelerator HaiJin(HuazhongUniversityofScienceandTechnology),PengchengYao(HuazhongUniversityofScienceandTechnology),XiaofeiLiao(HuazhongUniversityofScienceandTechnology),LongZheng(HuazhongUniversityofScienceandTechnology),XianliangLi(HuazhongUniversityofScienceandTechnology)
TowardsaRISCFrameworkforEfficientContextualizationinIoT DimitriosGeorgakopoulos(SwinburneUniversity),AliYavari(RMITUniversity),PremPrakashJayaraman(SwinburneUniversity),RajivRanjan(NewcastleUniversity)
TheFutureoftheSemanticWeb:PrototypesonaGlobalDistributedFilesystem MichaelCochez(Fraunhofer-FIT),DominikHüser(RWTHAachenUniversity),StefanDecker(RWTHAachen)
OnBroadBigData SteffenStaab(InstitutWeST,UniversityKoblenz-LandauandWAIS,UniversityofSouthampton)
![Page 46: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/46.jpg)
46
Research13:DistributedAlgorithmsandTheoryIILocation:SalonII
SessionChair:AzizMohaisen(StateUniversityofNewYorkatBuffalo)
NetworkedStochasticMulti-ArmedBanditswithCombinatorialStrategies ShaojieTang(UniversityofTexasatDallas),YaqinZhou(SUTD),KaiHan(UniversityofScienceandTechnologyofChina),ZhaoZhang(ZhejiangNormalUniversity),JingYuan(UniversityofTexasatDallas),WeiliWu(UniversityofTexasatDallas)
ComputabilityofPerpetualExplorationinHighlyDynamicRings MarjorieBournat(UPMCSorbonneUniversités),SwanDubois(UPMCSorbonneUniversités),FranckPetit(UPMCSorbonneUniversités)
LocallySelf-AdjustingSkipGraphs SikderHuq(TheUniversityofIowa),SukumarGhosh(TheUniversityofIowa)
OnlinetoOfflineBusiness:UrbanTaxiDispatchingwithPassenger-DriverMatchingStability HuanyangZheng(TempleUniversity),JieWu(TempleUniversity)
AnOptimizationFrameworkForOnlineRide-sharingMarkets YongzhengJia(TsinghuaUniversity),WeiXu(TsinghuaUniversity),XueLiu(McGillUniversity)
FastandAccurateTrackingofPopulationDynamicsinRFIDSystems MuhammadShahzad(NorthCarolinaStateUniversity),AlexLiu(MichiganStateUniversity)
ShortPaper5:DistributedBigDataSystemsandAnalyticsLocation:Nashville
SessionChair:VarunSoundararajan(Google)
TowardsMultilingualAutomatedClassificationSystems AibekMusaev(UniversityofAlabama),CaltonPu(GeorgiaInstituteofTechnology)
TheJointEffectsofTweetContentSimilarityandTweetInteractionsforTopicDerivation RobertusNugroho(MacquarieUniversity),WeiliangZhao(MacquarieUniversity),JianYang(MacquarieUniversity),CecileParis(CSIRO–ICTCentre),SuryaNepal(CSIRO)
Timed-releaseofSelf-emergingDatausingDistributedHashTables ChaoLi(UniversityofPittsburgh),BalajiPalanisamy(UniversityofPittsburgh)
CachingforPatternMatchingQueriesinTimeEvolvingGraphs:ChallengesandApproaches MuhammadNisar(UniversityofGeorgia),SaharVoghoei(UniversityofGeorgia),LakshmishRamaswamy(UniversityofGeorgia)
GraphA:AdaptivePartitioningforNaturalGraphs DongshengLi(NationalUniversityofDefenseTechnology),ChengfeiZhang(NationalUniversityofDefenseTechnology),JinyanWang(NationalUniversityofDefenseTechnology),ZhaoningZhang(NationalUniversityofDefenseTechnology),YimingZhang(NationalUniversityofDefenseTechnology)
ParallelAlgorithmforCoreMaintenanceinDynamicGraphs NaWang(HuazhongUniversityofScienceandTechnology),DongxiaoYu(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),ChenQian(HuazhongUniversityofScienceandTechnology),XiaXie(HuazhongUniversityofScienceandTechnology),Qiang-ShengHua(HuazhongUniversityofScienceandTechnology)
DHCRF:ADistributedConditionalRandomFieldAlgorithmonHeterogeneousCPU-GPUClusterforBigData AiWei(HunanUniversity),LiKenli(HunanUniversity),ChenCen(HunanUniversity),PengJiwu(HunanUniversity),LiKeqin(HunanUniversity)
TowardsNewAbstractionsforImplementingQuorum-basedSystems TormodErevikLea(UniversityofStavanger),LeanderJehl(UniversityofStavanger),HeinMeling(UniversityofStavanger)
SelectiveTrafficOffloadingOntheFly:aMachineLearningApproach ZaiyangTang(HuazhongUniversityofScienceandTechnology),PengLi(TheUniversityofAizu),SongGuo(TheHongKongPolytechnicUniversity),XiaofeiLiao(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),DaqingZhang(InstitutMines-Telecom,TelecomSudParis)
AFastHeuristicAttributeReductionAlgorithmusingSpark MinchengChen(WuhanUniversityofTechnology),JinglingYuan(WuhanUniversityofTechnology),LinLi(WuhanUniversityofTechnology),DonglingLiu(WuhanUniversityofTechnology),TaoLi(UniversityofFlorida)
ProfilingUsersbyModelingWebTransactions
![Page 47: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/47.jpg)
47
RadekTomsu(AaltoUniversity),SamuelMarchal(AaltoUniversity),N.Asokan(AaltoUniversity)
JeCache:Just-EnoughDataCachingwithJust-in-TimePrefetchingforBigDataApplications YifengLuo(FudanUniversity),JiaShi(FudanUniversity),ShuigengZhou(FudanUniversity)
Demo1-4Location:Charleston1and2
SessionChairs:GeraldFLofstead(SandiaNationalLab)andBalajiPalanisamy(UniversityofPittsburg)
Demo1:DistributedApplicationsCluster
LITMUS:TowardsMultilingualReportingofLandslides AibekMusaev(UniversityofAlabama),QixuanHou(GeorgiaInstituteofTechnology),YangYang(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology)
Pythia:ASystemforOnlineTopicDiscoveryofSocialMediaPosts IoulianaLitou(AthensUniversityofEconomicsandBusiness),VanaKalogeraki(AthensUniversityofEconomicsandBusiness)
Data-drivenSerendipityNavigationinUrbanPlaces XiaoyuGe(UniversityofPittsburgh),AmeyaDaphalapurkar(UniversityofPittsburgh),ManaliShimpi(UniversityofPittsburgh),KohliDarpun(UniversityofPittsburgh),KonstantinosPelechrinis(UniversityofPittsburgh),PanosChrysanthis(UniversityofPittsburgh),DemetriosZeinalipour-Yazti(MaxPlanckInstituteforInformaticsandUniversityofCyprus)
TowardAnIntegratedApproachtoLocalizingFailuresinCommunityWaterNetworks(DEMO) QingHan(UniversityofCaliforniaIrvine),PhuNguyen(UniversityofCaliforniaIrvine),RonaldT.Eguchi(ImageCat,Inc.),Kuo-LinHsu(UniversityofCaliforniaIrvine),NaliniVenkatasubramanian(UniversityofCaliforniaIrvine)
Demo2:SecurityandPrivacyCluster
PrivateGraph:ACloud-CentricSystemforPrivacy-PreservingSpectralAnalysisofLargeEncryptedGraphs SagarSharma(WrightStateUniversity),KekeChen(WrightStateUniversity)
IoTSentinelDemo:AutomatedDevice-TypeIdentificationforSecurityEnforcementinIoT MarkusMiettinen(TechnischeUniversitatDarmstadt),SamuelMarchal(AaltoUniversity),IbbadHafeez(universityofhelsinki),TommasoFrassetto(TechnischeUniversitatDarmstadt),N.Asokan(AaltoUniversity),Ahmad-RezaSadeghi(TechnischeUniversitatDarmstadt),SasuTarkoma(UniversityofHelsinki)
RogueAccessPointDetectorUsingCharacteristicsofChannelOverlappingin802.11nRhonghoJang(INHAuniversityofKorea),JeonilKang(INHAuniversity),AzizMohaisen(SUNYBuffalo),DaehunNyang(DepartmentofComputerandInformationEngineering,INHAUniversity,Incheon,Korea)
ReverseCloak:AReversibleMulti-levelLocationPrivacyProtectionSystem ChaoLi(UniversityofPittsburgh),BalajiPalanisamy(UniversityofPittsburgh),AravindKalaivanan(UniversityofPittsburgh),SriramRaghunathan(UniversityofPittsburgh)
Demo3:CloudsandVirtualizationCluster
Hopsworks:ImprovingUserExperienceandDevelopmentonHadoopwithScalable,StronglyConsistentMetadata MahmoudIsmail(KTH-RoyalInstituteofTechnology),ErmiasGebremeskel(RISESICS),TheofilosKakantousis(RISESICS),GautierBerthou(RISESICS),JimDowling(KTH-RoyalInstituteofTechnology)
IsolationinDockerthroughLayerEncryption IoannisGiannakopoulos(NationalTechnicalUniversityofAthens),KonstantinosPapazafeiropoulos(NationalTechnicalUniversityofAthens),KaterinaDoka(NationalTechnicalUniversityofAthens),NectariosKoziris(NationalTechnicalUniversityofAthens)
Dela-SharingLargeDatasetsbetweenHadoopClusters AlexandruA.Ormenisan(KTHRoyalInstituteofTechnology),JimDowling(KTHRoyalInstituteofTechnology)
InVivoEvaluationoftheSecureOpportunisticSchemesMiddlewareusingaDelayTolerantSocialNetworkCoreyE.Baker(UniversityofCaliforniaSanDiego),AllenStarke(UniversityofFlorida),TanishaG.Hill-Jarrett(UniversityofFlorida),JaniseMcNair(UniversityofFlorida)
Demo4:DistributedSystemsandNetworkingCluster
ScalingandLoadTestingLocation-basedPublishandSubscribe BertilChapuis(UniversityofLausanne),BenoîtGarbinato(UniversityofLausanne)
![Page 48: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/48.jpg)
48
ADistributedEvent-centricCollaborativeWorkflowsDevelopmentSystemforIoTApplication YongyangCheng(BUPT),ShuaiZhao(BUPT),BoCheng(BUPT),ShouluHou(BUPT),XiuleiZhang(BUPT),JunliangChen(BUPT)
IncentiveMechanismforData-CentricMessageDeliveryinDelayTolerantNetworks HimanshuJethawa(MissouriUniversityofScienceandTechnology),SanjayMadria(MissouriUniversityofScienceandTechnology)
PerformanceOfCognitiveWirelessChargerForNear-FieldWirelessChargingSang-YoonChang(UniversityofColorado,ColoradoSpringsandAdvancedDigitalSciencesCenter),SristiLakshmiSravanaKumar(AdvancedDigitalSciencesCenter),Yih-ChunHu(UniversityofIllinoisatUrbana-Champaign)
19:00-21:00 Wednesday, June 7, 2017 OrganizationEvent(InvitationOnly)
![Page 49: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/49.jpg)
49
Day 4 – Thursday, June 8, 2017
7:30-8:30 Thursday, June 8, 2017 ContinentalBreakfastLocation:Foyer
8:30-9:30 Thursday, June 8, 2017 Keynote3byKennethCalvert(NationalScienceFoundation)Location:PhoenixBallroom
SessionChair:CaltonPu(GeorgiaInstituteofTechnology)
9:30-10:00 Thursday, June 8, 2017 CoffeeBreakLocation:Foyer
10:00-12:00 Thursday, June 8, 2017 Tutorial2:SensorCloud:ACloudofSensorNetworksLocation:Charleston1
SanjayMadria(MissouriUniversityofScienceandTechnology)
Research14:MobileandWirelessComputingSystemsIILocation:SalonI
SessionChair:CarleeJoe-Wong(CarnegieMellonUniversity)
RobustIndoorWirelessLocalizationUsingSparseRecovery WeiGong(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity)
Max-MinFairResourceAllocationinHetNets:DistributedAlgorithmsandHybridArchitecture EhsanAryafar(PortlandStateUniversity),AlirezaKeshavarz-Haddad(ShirazUniversity),CarleeJoe-Wong(CarnegieMellonUniversity),MungChiang(PrincetonUniversity)
OptimizationofFull-ViewBarrierCoveragewithRotatableCameraSensors XiaofengGao(ShanghaiJiaoTongUniversity),RuiYang(UniversityofIllinoisUrbana-Champaign),FanWu(ShanghaiJiaoTongUniversity),GuihaiChen(ShanghaiJiaoTongUniversity),JingguangZhou(ShanghaiJiaoTongUniversity)
CommunicationthroughSymbolSilence:TowardsFreeControlMessagesinIndoorWLANs BingFeng(UniversityofScienceandTechnologyofChina),JianqingLiu(UniversityofFlorida),ChiZhang(UniversityofScienceandTechnologyofChina),YuguangFang(UniversityofFlorida)
Secureconnectivityofwirelesssensornetworksunderkeypredistributionwithon/offchannels JunZhao(CarnegieMellonUniversity)
iUpdater:LowCostRSSFingerprintsUpdatingforDevice-freeLocalization LiqiongChang(NorthwestUniversity),JieXiong(SingaporeManagementUniversity),YuWang(UniversityofNorthCarolinaatCharlotte),XiaojiangChen(NorthwestUniversity),JunhaoHu(NorthwestUniversity),FangDingyi(NorthwestUniversity)
Research15:SocialNetworksandCrowdsourcingLocation:Columbia
SessionChair:TingWang(LehighUniversity)
InfluenceMaximizationinaManyCascadesWorldIoulianaLitou(AUEB),VanaKalogeraki(AUEB),DimitriosGunopulos(UoA)
Expertise-AwareTruthAnalysisandTaskAllocationinMobileCrowdsourcingXiaomeiZhang(UniversityofSouthCarolinaBeaufort),YiboWu(PennsylvaniaStateUniversity),LifuHuang(RensselaerPolytechnicInstitute),HengJi(RensselaerPolytechnicInstitute),GuohongCao(PennsylvaniaStateUniversity)
![Page 50: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/50.jpg)
50
MeLoDy:ALong-termDynamicQuality-awareIncentiveMechanismforCrowdsourcing HongweiWang(ShanghaiJiaoTongUniversity),SongGuo(TheHongKongPolytechnicUniversity),JiannongCao(TheHongKongPolytechnicUniversity),MinyiGuo(ShanghaiJiaoTongUniversity)
TheStrongLinkGraphforEnhancingSybilDefenses SuhendryEffendy(NationalUniversityofSingapore),RolandYap(NationalUniversityofSingapore)
MechanismDesignforMobileCrowdsensingwithExecutionUncertainty ZhenzheZheng(ShanghaiJiaoTongUniversity),ZhaoxiongYang(ShanghaiJiaoTongUniversity),FanWu(ShanghaiJiaoTongUniversity),GuihaiChen(ShanghaiJiaoTongUniversity)
TowardsScalableandDynamicSocialSensingUsingADistributedComputingFramework DanielZhang(UniversityofNotreDame),ChaoZheng(UniversityofNotreDame),DongWang(UniversityofNotreDame),DougThain(UniversityofNotreDame),XinMu(UniversityofNotreDame),GregMadey(UniversityofNotreDame),ChaoHuang(UniversityofNotreDame)
Application9:DistributedSystemsandApplicationsLocation:Atlanta
SessionChair:XueyanTang(NanyangTechnologicalUniversity)
SpecifyingaDistributedSnapshotAlgorithmasaMeta-programandModelCheckingitatMeta-level HaThiThuDoan(JapanAdvancedInstituteofScienceandTechnology),FrancoisBonnet(OsakaUniversity),KazuhiroOgata(JapanAdvancedInstituteofScienceandTechnology)
Self-EvolvingSubscriptionsforContent-BasedPublish/SubscribeSystems CesarCañas(McGillUniversity),KaiwenZhang(TechnicalUniversityofMunich),BettinaKemme(McGillUniversity),JörgKienzle(McGillUniversity),Hans-ArnoJacobsen(TechnicalUniversityofMunich)
ScalableRoutingforTopic-basedPublish/SubscribeSystemsunderFluctuations VolkerTurau(HamburgUniversityofTechnology),GerrySiegemund(HamburgUniversityofTechnology)
OPPay:DesignandImplementationofAPaymentSystemforOpportunisticDataServices FengruiShi(ImperialCollegeLondon),ZhijinQin(ImperialCollegeLondon),JulieMcCann(ImperialCollegeLondon)
OptimalResourceAllocationforMulti-userVideoStreamingovermmWaveNetworksZhifengHe(AuburnUniversity),ShiwenMao(AuburnUniversity)
AMulti-AgentParallelApporachtoAnalyzingLargeClimateDataSets JasonWoodring(UniversityofWashingtonBothell),MatthewSell(UniversityofWashingtonBothell),MunehiroFukuda(UniversityofWashingtonBothell),HazelineAsuncion(UniversityofWashingtonBothell),EricSalathe(UniversityofWashingtonBothell)
Application10:DistributedSystemsandServicesLocation:Columbia
SessionChair:SangeethaSeshadri(IBMAlmadenResearchCenter)
EnergyProportionalServers:WhereAreWein2016? CongfengJiang(HangzhouDianziUniversity),YumeiWang(HangzhouDianziUniversity),DongyangOu(HangzhouDianziUniversity),BingLuo(WayneStateUniversity),WeisongShi(WayneStateUniversity)
AreHTTP/2ServersReadyYet? MuhuiJiang(TheHongKongPolytechnicUniversity),XiapuLuo(TheHongKongPolytechnicUniversity),TungngaiMiu(NexusguardLimited),ShengtuoHu(TheHongKongPolytechnicUniversity),WeixiongRao(TongjiUniversity)
DataIntegrityforCollaborativeApplicationsoverHostedServices ErtemEsiner(NanyangTechnologicalUniversity),AnwitamanDatta(NanyangTechnologicalUniversity)
VirtualMachinePowerAccountingwithShapleyValue WeixiangJiang(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),GuomingTang(UniversityofVictoria),KuiWu(UniversityofVictoria),HaiJin(HuazhongUniversityofScience&Technology)
AVersatilePlatformforMobileDataGatheringExperimentsinWirelessSensorNetworks JiLi(StonyBrookUniversity),CongWang(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity)
OnDirectionalNeighborDiscoveryinmmWaveNetworks YuWang(AuburnUniversity),ShiwenMao(AuburnUniversity),TheodoreS.Rappaport(NewYorkUniversity)
![Page 51: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/51.jpg)
51
ShortPaper6:Security,Privacy,Trust,andFaultToleranceinDistributedSystemsLocation:Nashville
SessionChair:BalajiPalanisamy(UniversityofPittsburgh)
ProximityAwarenessApproachtoEnhancePropagationDelayontheBitcoinPeer-to-PeerNetwork MuntadherFadhilSallal(UniversityofPortsmouth),GarethOwenson(UniversityofPortsmouth),MoAdda(UniversityofPortsmouth)
CatchMeIfYouCan:DetectingCompromisedUsersThroughPartialObservationonNetworksDerekWang(DeakinUniversity),ShengWen(DeakinUniversity),JunZhang(DeakinUniversity),SuryaNepal(Data61),YangXiang(DeakinUniversity),WanleiZhou(DeakinUniversity)
LocationPrivacyBreach:AppsAreWatchingYouinBackground DachuanLiu(CollegeofWilliam&Mary),XingGao(CollegeofWilliam&Mary),HainingWang(UniversityofDelaware)
AndroidMalwareDetectionusingComplex-Flows FengShen(SUNYBuffalo),JustinDelVecchio(SUNYBuffalo),AzizMohaisen(SUNYBuffalo),StevenY.Ko(SUNYBuffalo),LukaszZiarek(SUNYBuffalo)
PrivacyImplicationsofDNSSECLook-asideValidation AzizMohaisen(SUNYBuffalo),ZhongshuGu(IBMResearch),KuiRen(SUNYBuffalo)
FlipNet:ModelingCovertandPersistentAttacksonNetworkedResources SudipSaha(VirginiaPolytechnicInstituteandStateUniversity),AnilVullikanti(VirginiaPolytechnicInstituteandStateUniversity),MahanteshHalappanavar(PacificNorthwestNationalLab)
UnderstandingtheMarket-levelandNetwork-levelBehaviorsoftheAndroidMalwareEcosystem ChaoYang(Niara,Inc.),JialongZhang(IBMResearch),GuofeiGu(TexasA&MUniversity)
EnGarde:Mutually-TrustedInspectionofSGXEnclaves. HaiNguyen(RutgersUniversity),VinodGanapathy(RutgersUniversity)
TruthfulOnlineAuctionforCloudInstanceSubletting YifeiZhu(SimonFraserUniversity),SilveryFu(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity),YongCui(TsinghuaUniversity)
OnthePowerofWeakerPairwiseInteraction:Fault-TolerantSimulationofPopulationProtocols GiuseppeAntonioDiLuna(LaSapienza),PaolaFlocchini(UniversityofOttawa),TaisukeIzumi(NagoyaInstituteofTechnology),TomokoIzumi(CollegeofInformationScienceandEngineering),NicolaSantoro(CarletonUniversity),GiovanniViglietta(UniversityofOttawa)
DistributedFaultTolerantLinearSystemSolversbasedonErasureCoding XuejiaoKang(PurdueUniversity--WestLafayette),DavidF.Gleich(PurdueUniversity--WestLafayette),AhmedSameh(PurdueUniversity--WestLafayette),AnanthGrama(PurdueUniversity--WestLafayette)
PreservingIncumbentUsers’PrivacyinExclusion-Zone-BasedSpectrumAccessSystems YanzhiDou(VirginiaTech),HeLi(VirginiaTech),KexiongZeng(VirginiaTech),JinshanLiu(VirginiaTech),YalingYang(VirginiaTech),BoGao(ChineseAcademyofSciences),KuiRen(SUNYBuffalo)
12:00-13:30 Thursday, June 8, 2017 BusinessLunchincludingAwardsandICDCS2018AnnouncementsLocation:PhoenixBallroom
13:30-15:30 Thursday, June 8, 2017 Tutorial2:SensorCloud:ACloudofSensorNetworks(until14:30)Location:Charleston1
SanjayMadria(MissouriUniversityofScienceandTechnology) PlenaryPanelontheConvergenceofBigData,IoT/CPS,andSCC(Smart&ConnectedCommunities)Location:PhoenixBallroom
ThekeywordsofBigData,InternetofThings(IoT),Cyber-PhysicalSystems(CPS),andSmart&ConnectedCommunities(a.k.a.SmartCitiesandSmartPlanet)havebeenaroundforquiteafewyears.Theystartedfromdifferentdisciplines:
![Page 52: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/52.jpg)
52
• Bigdataoriginatedfromscientificsensors(e.g.,satellites)andenterpriseapplications(e.g.,Walmartcashregisters).
• IoT/CPSstartedfromwirelesssensornetworksandreal-timesystems(e.g.,theCPSweekseriesofconferences)• SmartPlanet,SmartCities,andSCCstartedfromAutonomicComputing(withthesekeywordspopularized
initiallybyIBM)
Theseresearchfieldsandapplicationareashavebeenexpandingtremendouslyinrecentyears,withsignificantandgrowingoverlapsamongthem.Webelievetheseoverlapsrepresenthugeopportunitiesforinnovativeresearchandpotentialsocietalimpact.Atthesametime,theyalsocreateunprecedentedresearchchallengesthatsurpassthetraditionalconfinesofeachoriginaldiscipline.
Assimpleexamples,bigdataanalyticsandreal-timeIoTsensor-baseddecisionmaking(individually)havebeenthefrontrunnersmartapplicationsshowcasedbySmartCityprojects.However,theircombinationhasseenlimitedintegrationoftechniquesfrombothfields.Anexampleofoverlappingareasisstreaminganalyticsofreal-timeIoTsensors,whichtypicallyusebigdatatoolsatfinetimegranularity(e.g.,ApacheSparkandStorm)withlingeringlimitationsinreal-timeguaranteesandmachinelearningcapabilitiesneededforsmartapplications.Aconcreteexampleofchallengeapplicationsistheautomatedreal-timetrackingofmovingentitiesacrossnetworksofvideocameras.Althoughsuchtrackingisdoneroutinelybyhumans,itsautomationrequiressignificantadvancesintheintegrationofvision/machinelearningtools(objectrecognitionfromvideoimages)withstreaminganalytics(parallelprocessingofvideoimagesfromrelatedcameras).
Thepanelwilldiscusstheoverlappingareasamongbigdata,IoT/CPS,andSCC/SmartCities,withemphasisontheresearchopportunitiesandtechnicalchallenges.
PanelModerator:CaltonPu
CaltonPuCalton'sresearchinterestsareintheareasofservicecomputing,distributedandcloudcomputing,integrationandveracityofbigdata.Hiscurrentprojectsincludecloudcomputing(Elba)andbigdata(GRAIT-DM)research.Usingexperimentaldatafromrealisticbenchmarks,theElbaprojectstudiestheinterestingphenomenasuchasveryshortbottlenecksthathavelargeimpactonn-tiersystemresponsetime.TheGRAIT-DMprojectcollectsrealworlddatafromsocialsensors(e.g.,TwitterandYouTube)andphysicalsensors(e.g.,USGSGSNandNASATRMM)todetectphysicaleventsandmanagereal-timeinformationonthem.ThesponsorsforCaltonPu'sresearchincludebothgovernmentfundingagenciessuchasNSF,andcompaniesfromindustrysuchasHP,Fujitsu,IBM,andIntel.Heisaco-directorofCenterforExperimentalResearchinComputerSystems(CERCS),andaffiliatefacultyofInstituteforInformationSecurityandPrivacy(IISP)atGeorgiaTech.HeisalsothedirectorofRCNonBigDataforSmartCities,withcollaborationsaroundtheworld.Positionsavailable:GeorgiaTechisrecruitinggoodgraduatestudents.
Panelists
![Page 53: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/53.jpg)
53
KarlAbererKarlAbererreceivedhisPhDinmathematicsin1991fromtheETHZürich.From1991to1992hewaspostdoctoralfellowattheInternationalComputerScienceInstitute(ICSI)attheUniversityofCalifornia,Berkeley.In1992,hejoinedtheIntegratedPublicationandInformationSystemsinstitute(IPSI)ofGMDinGermany,wherehewasleadingtheresearchdivisionOpenAdaptiveInformationManagementSystems.In2000hejoinedEPFLasfullprofessor.Since2005heisthedirectoroftheSwissNationalResearchCenterforMobileInformationandCommunicationSystems(NCCR-MICS,www.mics.ch).HeismemberoftheeditorialboardsofVLDBJournal,ACMTransactiononAutonomousandAdaptiveSystemsandWorldWideWebJournal.HehasbeenconsultingfortheSwissgovernmentinresearchandsciencepolicyasamemberoftheSwissResearchandTechnologyCouncil(SWTR)from2003-2011.
SrinivasAluruSrinivasAluruisaprofessorintheSchoolofComputationalScienceandEngineeringwithintheCollegeofComputingatGeorgiaInstituteofTechnology.Heco-leadstheGeorgiaTechStrategicInitiativeinDataEngineeringandScience.Heconductsresearchinhighperformancecomputing,bioinformaticsandsystemsbiology,combinatorialscientificcomputing,andappliedalgorithms.Hepioneeredthedevelopmentofparallelmethodsincomputationalbiology,andcontributedtotheassemblyandanalysisofcomplexplantgenomes.Hisgroupiscurrentlyfocusedondevelopingbioinformaticsmethodsforhigh-throughputDNAsequencing,particularlyerrorcorrectionandgenomeassembly.Insystemsbiology,hisgroupisworkingonnetworkinferencemethodsusingmutualinformationandBayesianapproaches,andnetworkanalysistechniquestofurthertheknowledgeofpartiallycharacterizedpathways.HiscontributionsinscientificcomputinglieinparallelFastMultipoleMethod,domaindecompositionmethods,spatialdatastructures,andapplicationsincomputationalelectromagneticsandmaterialsinformatics.AluruisaFellowoftheAmericanAssociationfortheAdvancementofScience(AAAS)andtheInstituteforElectricalandElectronicEngineers(IEEE).HeisarecipientoftheNSFCareeraward(1997),IBMfacultyaward(2002),andSwarnajayantifellowshipfromtheGovernmentofIndia(2007).HeservesontheeditorialboardsoftheIEEETransactionsonParallelandDistributedSystems,theJournalofParallelandDistributedComputing,andtheInternationalJournalofDataMiningandBioinformatics.
KennethCalvertKenCalvertisDivisionDirectorforComputerandNetworkSystemsintheComputerandInformationScienceandEngineering(CISE)DirectorateattheNationalScienceFoundation.HeisonrotationfromtheUniversityofKentucky,whereheisGartnerGroupProfessorinNetworkEngineeringintheDepartmentofComputerScience.Hisresearchdealswiththedesignandimplementationofadvancednetworkprotocolsandservices,withparticularinterestinroutingandincentivesinfuturenetworkarchitectures.HereceivedhisPh.D.incomputersciencefromtheUniversityofTexasatAustin.HeholdsaM.S.incomputersciencefromStanfordUniversityandaB.S.incomputerscienceandengineeringfromtheMassachusettsInstituteofTechnology.PriortohisappointmentattheUniversityofKentucky,hewasaMemberoftheTechnicalStaffatBellLaboratoriesinHolmdel,NJ,andservedonthefacultyintheCollegeofComputing
![Page 54: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/54.jpg)
54
attheGeorgiaInstituteofTechnology.HeisanIEEEFellowandamemberoftheACM.
C.MohanDr.C.MohanhasbeenanIBMresearcherfor35yearsinthedatabasearea,impactingnumerousIBMandnon-IBMproducts,theresearchandacademiccommunities,andstandards,especiallywithhisinventionoftheARIESfamilyofdatabaselockingandrecoveryalgorithms,andthePresumedAbortcommitprotocol.ThisIBM(1997),andACM/IEEE(2002)FellowhasalsoservedastheIBMIndiaChiefScientistfor3years(2006-2009).InadditiontoreceivingtheACMSIGMODInnovationAward(1996),theVLDB10YearBestPaperAward(1999)andnumerousIBMawards,MohanwaselectedtotheUSandIndianNationalAcademiesofEngineering(2009),andwasnamedanIBMMasterInventor(1997).ThisDistinguishedAlumnusofIITMadras(1977)receivedhisPhDattheUniversityofTexasatAustin(1981).Heisaninventorof50patents.HeiscurrentlyfocusedonBigData,HTAPandBlockchaintechnologies.In2016,hewasnamedaDistinguishedVisitingProfessorofChina’sprestigiousTsinghuaUniversity.HehasservedontheadvisoryboardofIEEESpectrum,andonnumerousconferenceandjournalboards.MohanisafrequentspeakerinNorthAmerica,EuropeandIndia,andhasgiventalksin40countries.Heisveryactiveonsocialmediaandhasahugenetworkoffollowers.MoreinformationcouldbefoundintheWikipediapageathttp://bit.ly/CMwIkP
ManishParasharManishParasharisDistinguishedProfessorofComputerScienceatRutgers,TheStateUniversityofNewJerseyUniversity.HeisalsothefoundingDirectoroftheRutgersDiscoveryInformaticsInstitute(RDI2)andTheAppliedSoftwareSystemsLaboratory(TASSL),FullMember(ClinicalInvestigationsandPrecisionTherapeuticsProgram)oftheRutgersCancerInstituteofNewJersey,andisAssociateDirectorattheRutgersCenterforInformationAssurance(RUCIA).HealsohasaJointFacultyAppointmentwithOakRidgeNationalLaboratory(ORNL),andisVisitingProfessorintheFacultyofBusiness,Computing&Law,UniversityofDerby,UK.Heco-foundedandwasCo-DirectoroftheCloudandAutonomicComputingCenter(CAC)NSFIUCRCatRutgers(CAC@Rutgers)between2008and2013.AtRutgers,heled(withProf.H.Berman)thestrategicplanningeffortsinResearchComputingandservedastheInterimAssociateVicePresidentofResearchComputingbetween2015–2016tooverseetheestablishmentoftheRutgersOfficeofAdvancedResearchComputing(OARC).HeisalsocurrentlytheLeadPIforCyberinfrastructurefortheNSFOceanObservatoriesInitiative.
![Page 55: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/55.jpg)
55
Research Track Paper Abstracts
Research1:DistributedFaultToleranceandDependabilityTimely,Reliable,andCost-EffectiveInternetTransportServiceusingDisseminationGraphs AmyBabay(JohnsHopkinsUniversity),EmilyWagner(JohnsHopkinsUniversity,LTNGlobalCommunications),MichaelDinitz(JohnsHopkinsUniversity),YairAmir(JohnsHopkinsUniversity,LTNGlobalCommunications)
Emergingapplicationssuchasremotemanipulationandremoteroboticsurgeryrequirecommunicationthatisbothtimelyandreliable,buttheInternetnativelysupportsonlycommunicationthatiseithercompletelyreliablewithnotimelinessguarantees(e.g.TCP)ortimelywithbest-effortreliability(e.g.UDP).Wepresentanoverlaytransportservicethatcanprovidehighlyreliablecommunicationwhilemeetingstringenttimelinessguarantees(e.g.130msround-triplatencyacrosstheUS)overtheInternet.Toenableroutingschemesthatcansupportthenecessarytimelinessandreliability,weintroducedisseminationgraphs,providingaunifiedframeworkforspecifyingroutingschemesrangingfromasinglepath,tomultipledisjointpaths,toarbitrarygraphs.Weconductanextensiveanalysisofrealworldnetworkdata,findingthataroutingapproachusingtwodisjointpathsperformswellinmostcases,andthatcaseswheretwodisjointpathsdonotperformwelltypicallyinvolveproblemsaroundasourceordestination.Basedonthisanalysis,wedevelopatimelydissemination-graph-basedroutingmethodthatcanaddtargetedredundancyinproblematicareasofthenetwork.Thisapproachcancoverover99%oftheperformancegapbetweenatraditionalsingle-pathapproachandanoptimal(butprohibitivelyexpensive)scheme,whiletwodynamicdisjointpathscoverabout70%ofthisgap,andtwostaticdisjointpathscoverabout45%.Thisperformanceimprovementisobtainedatacostincreaseofabout2%overtwodisjointpaths.
Pronto:EfficientTestPacketGenerationforDynamicNetworkDataPlanes YuZhao(UniversityofKentucky),HuazheWang(UniversityofCaliforniaatSantaCruz),XinLin(UniversityofCaliforniaatSantaCruz),TingtingYu(UniverstiyofKentucky),ChenQian(UniversityofCaliforniaatSantaCruz)
Computernetworksarebecomingincreasinglycomplextodayandthuspronetovariousnetworkfaults.Traditionaltestingtools(e.g.,ping,traceroute)thatofteninvolvesubstantialmanualefforttouncoverfaultsareinefficient.Thispaperfocusesonfaultdetectionofthenetworkdataplaneusingtestpackets.Existingsolutionsoftestpacketgenerationeithertakeverylongtime(e.g.,morethanonehour)tocompleteorgeneratetoomanytestpacketsthatmayhurtregulartraffic.Inthispaper,wepresentPronto,anautomatedtestpacketgenerationtoolthatgeneratestestpacketstoexercisedataplanerulesintheentirenetworkinashorttime(e.g.,severalseconds)andcanquicklyreacttorulechangesduetonetworkdynamics.Inaddition,Prontominimizesthenumberoftestpacketsbyallowingapackettotestmultiplerulesatdifferentswitches.TheperformanceevaluationusingtworealnetworkdataplanerulesetsshowsthatProntoisfasterthanarecentlydevelopedtoolbymorethantwoordersofmagnitude.Prontocanupdatetheprobesforrulechangesusinglessthan1mswhileexistingmethodshavenosuchupdatefunction.
Agar:ACachingSystemforErasure-CodedData RalucaHalalai(UniversityofNeuchâtel),PascalFelber(UniversityofNeuchâtel),Anne-MarieKermarrec(INRIA),FrançoisTaïani(IRISA)
Erasurecodingisanestablisheddataprotectionmechanism.Itprovideshighresiliencywithlowstorageoverhead,whichmakesitveryattractivetostoragesystemsdevelopers.Unfortunately,whenusedinadistributedsetting,erasurecodinghampersastoragesystemsperformance,becauseitrequiresclientstocontactseveral,possiblyremotesitesinordertoretrievetheirdata.Thishashinderedtheadoptionoferasurecodinginpractice,limitingitsusetocold,archivaldata.Recentresearchshowedthatitisfeasibletouseerasurecodingforhotdataaswell,thusopeningupnewperspectivesforimprovingerasurecodedstoragesystems.Inthispaper,weaddresstheproblemofminimizingaccesslatencyinerasure-codedstorage.WeproposeAgaranovelcachingsystemtailoredforerasure-codedcontent.Agaroptimizesthecontentsofthecachebasedonliveinformationregardingdatapopularityandaccesslatencytodifferentdatastoragesites.Oursystemadaptsadynamicprogrammingalgorithmtooptimizethechoiceofdatablocksthatarecached,usinganapproachakintoKnapsackalgorithms.WecompareAgartotheclassicalLeastRecentlyUsedandLeastFrequentlyUsedcacheevictionpolicies,whilevaryingtheamountofdatacachedbetweenadatachunkandawholereplicaoftheobject.WeshowthatAgarcanachieve16%to41%lowerlatencythansystemsthatuseclassicalcachingpolicies.
Highperformancerecoveryforparallelstatemachinereplication OdoricoMendizabal(FURG),FernandoLuisDotti(PUCRS),FernandoPedone(UniversityofLugano)
Statemachinereplicationisafundamentalapproachtohighavailability.Despitethevastliteratureonthetopic,relativelyfewstudieshaveconsideredtheissuesinvolvedinrecoveringfaultyreplicas.Recoveringareplicarequires(a)retrievingandinstallinganup-to-datereplicacheckpoint,and(b)restoringandre-executingthelogofcommandsnotreflectedinthecheckpoint.Paralleltechniquestostatemachinereplicationrenderrecoveryparticularlychallengingsincethroughputundernormalexecution(i.e.,intheabsenceoffailures)isveryhigh.Consequently,thelogofcommandsthatneedtobeapplieduntilthereplicaisavailableistypicallylarge,whichdelaysrecovery.Inthispaper,wepresenttwotechniquestooptimizerecoveryinparallelstatemachinereplication.Thefirsttechniqueallowsnewcommandstoexecuteconcurrentlywiththeexecutionofloggedcommands,beforereplicasarecompletelyupdated.Thesecondtechniqueintroduceson-demandstaterecovery,whichallowssegmentsofacheckpointtoberecoveredconcurrently.
OnDataParallelismofErasureCodinginDistributedStorageSystems JunLi(UniversityofToronto),BaochunLi(UniversityofToronto)
Deployedinvariousdistributedstoragesystems,erasurecodinghasdemonstrateditsadvantagesoflowstorageoverheadandhighfailuretolerance.Typicallyinanerasurecodeddistributedstoragesystem,systematicmaximumdistanceseperable(MDS)codesarechosensincetheoptimalstorageoverheadcanbeachievedandmeanwhiledatacanbereaddirectlywithoutdecodingoperations.However,dataparallelismofexistingMDScodesislimited,becausewecanonlyreaddatafromsomespecificserversinparallelwithoutdecodingoperations.Inthispaper,weproposeCarouselcodes,designedtoallowdatatobereadfromanarbitrarynumberofserversinparallelwithoutdecoding,whilepreservingtheoptimalstorageoverheadofMDScodes.Furthermore,Carouselcodescanachievetheoptimalnetworktraffictoreconstructanunavailableblock.WehaveimplementedaprototypeofCarouselcodesonApacheHadoop.OurexperimentalresultshavedemonstratedthatCarouselcodescanmakeMapReducejobsfinishwithalmost50%lesstimeandreducedataaccesslatencysignificantly,withacomparablethroughputintheencodinganddecodingoperationsandnoadditionalsacrificeoffailuretoleranceorthenetworkoverheadtoreconstructunavailabledata.
MeteorShower:MinimizingRequestLatencyforMajorityQuorum-basedDataConsistencyAlgorithmsinMultipleDataCenters
![Page 56: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/56.jpg)
56
YingLiu(KTHRoyalInstituteofTechnology),XiGuan(KTHRoyalInstituteofTechnology),VladimirVlassov(KTHRoyalInstituteofTechnology),SeifHaridi(KTHRoyalInstituteofTechnology)
Withtheincreasingpopularityofservingandstoringdatainmultipledatacenters,weinvestigatetheefficiencyofmajorityquorum-baseddataconsistencyalgorithmsunderthisscenario.Becauseofthefailure-pronenatureofdistributedstoragesystems,majorityquorum-baseddataconsistencyalgorithmsbecomeoneofthemostwidelyadoptedapproaches.Inthispaper,weproposetheMeteorShowerframework,whichprovidesfaulttolerantread/writekey-valuestorageserviceacrossmultipledatacenterswithsequentialconsistencyguarantees.Amajorfeatureisthatmostreadoperationsareexecutedlocallywithinasingledatacenter.Thisresultsinloweringreadlatencyfromhundredsofmillisecondstotensofmilliseconds.ThedataconsistencyalgorithminMeteorShoweraugmentsmajorityquorum-basedalgorithms.Thus,itkeepsallthedesirablepropertiesofmajorityquorums,suchasfaulttolerance,balancedload,etc.AnimplementationofMeteorShowerontopofCassandraisdeployedandevaluatedinmultipledatacentersusingtheGoogleCloudPlatform.EvaluationsofMeteorShowerframeworkhaveshownthatitcanconsistentlyservereadrequestswithoutpayingthecommunicationdelaysamongreplicasmaintainedinmultipledatacenters.Asaresult,weareabletoimprovethelatencyofreadrequestsfromhundredsofmillisecondstotensofmillisecondswhileachievingthesamelatencyonwriterequestsandthesamefaulttoleranceguarantee.Thus,MeteorShowerisoptimizedforreadintensiveworkloads.
Research2:DistributedOperatingSystemsandMiddlewareLSbM-tree:Re-enablinghigh-speedcachinginDataManagementforMixedReadsandWrites DejunTeng(TheOhioStateUniversity),LeiGuo(Google),RubaoLee(TheOhioStateUniversity),FengChen(LouisianaStateUniversity),SiyuanMa(TheOhioStateUniversity),XiaodongZhang(TheOhioStateUniversity),YanfengZhang(NortheasternUniversity)
LSM-treehasbeenwidelyusedindatamanagementproductionsystemsforwrite-intensiveworkloads.However,asreadandwriteworkloadsco-existunderLSM-tree,dataaccessescanexperiencelonglatencyandlowthroughputduetotheinterferencestobuffercachingfromthecompaction,amajorandfrequentoperationinLSM-tree.Afteracompaction,theexistingdatablocksarereorganizedandwrittentootherlocationsondisks.Asaresult,therelateddatablocksthathavebeenloadedinthebuffercacheareinvalidatedsincetheirreferencingaddressesarechanged,causingseriousperformancedegradations.Inordertore-enablehigh-speedbuffercachingduringintensivewrites,weproposeLog-Structuredbuffered-Mergetree(simplifiedasLSbM-tree)byaddingacompactionbufferondisks,tominimizethecacheinvalidationsonbuffercachecausedbycompactions.Thecompactionbufferefficientlyandadaptivelymaintainsthefrequentlyvisiteddatasets.InLSbM,stronglocalityobjectscanbeeffectivelykeptinthebuffercachewithminimumorwithoutharmfulinvalidations.Withthehelpofasmallon-diskcompactionbuffer,LSbMachievesahighqueryperformancebyenablingeffectivebuffercaching,whileretainingallthemeritsofLSM-treeforwrite-intensivedataprocessing,andprovidinghighbandwidthofdisksforrangequeries.WehaveimplementedLSbMbasedonLevelDB.Weshowthatwithastandardbuffercacheandaharddisk,LSbMcanachieve2xperformanceimprovementoverLevelDB.WehavealsocomparedLSbMwithotherexistingsolutionstoshowitsstrongeffectiveness.
IncrementalTopologyTransformationforPublish/SubscribeSystemsUsingIntegerProgramming PooyaSalehi(TechnicalUniversityofMunich),KaiwenZhang(TechnicalUniversityofMunich),Hans-ArnoJacobsen(UniversityofToronto)
Distributedoverlay-basedpublish/subscribesystemsprovideaselective,scalable,anddecentralizedapproachtodatadissemination.Duetothedynamiccommunicationflowsbetweendataproducersandconsumers,theoverlaytopologyofsuchsystemscanbecomeinefficientovertimeandthereforerequiresadaptationtotheexistingload.Existingstudiesproposealgorithmstodesignoverlaytopologieswhichareoptimizedforspecificworkloads.However,theproblemofgeneratingaplantoincrementallytransformthecurrenttopologytoanoptimizedonehasbeenlargelyignored.Inthispaper,wepresentIPITT,anapproachbasedonintegerprogrammingfortheincrementaltopologytransformation(ITT)problem.Giventhecurrenttopologyandatargettopology,IPITTgeneratesatransformationplanwithaminimalnumberofstepsinordertolessenservicedisruption.Furthermore,weintroduceaplanexecutionmechanismandevaluateourapproachonanexistingpublish/subscribesystem.Basedonourevaluation,IPITTcanreduceplancomputationtimebyafactorof10andgeneratesplanswithanexecutiontimeupto55%shorterthanthoseofexistingapproaches.
milliScope:aFine-GrainedMonitoringFrameworkforPerformanceDebuggingofn-TierWebServices Chien-AnLai(GeorgiaInstituteofTechnology),JoshKimball(GeorgiaInstituteofTechnology),TaoZhu(GeorgiaInstituteofTechnology),QingyangWang(LouisianaStateUniversity),CaltonPu(GeorgiaInstituteofTechnology)
Moderndistributedsystemsareoftenconsideredtobeblack-boxesthatgreatlylimitthepotentialtounderstandbehaviorsatthelevelofdetailnecessarytodiagnosesomeofthemostimportanttypesofperformanceproblems.Recentlyresearchershavefoundabnormalresponsetimedelay,onetotwoorderofmagnitudelongertimethantheaverageresponsetime,existsinshortperiodandcauseseconomicallossforserviceproviders.Theseveryshortbottlenecksarehardtodetectduetoitsshortlivespananditsvarietyofpossiblereasons.Inthispaper,weproposemilliScope(mScope),thefirstmillisecond-granularitysoftware-basedresourceandeventmonitoringfordistributedsystemsthatachievesbothperformance,lowoverheadathighfrequency,andhighaccuracymatchedwithotherfirmwaremonitoringtool.Morespecifically,milliScopeisafine-grainedmonitoringframeworktocollaboratemultiplemScopeMonitorsforeventandresourcemonitoringtoreconstructtheflowofeachclientrequestandprofileexecutionperformanceinadistributedsystem.WeutilizetheresourcemScopeMonitorsforsystemresourcemonitoring,andwedevelopourowneventmScopeMonitorstoidentifytheexecutionboundaryinalightweight,preciseandsystematicmethodology.Thesemanticandsyntacticofthesemonitoringlogswitharbitraryformatsareenrichedbyourmulti-stagedatatransformationtool,mScopeDataTransformer,whichunifiesthediversemonitoringlogsintoadynamicdatawarehouse,mScopeDB,foradvancedanalysis.WeconductseveralillustrativescenariosinwhichmilliScopesuccessfullydiagnosestheresponsetimeanomaliescausedbyveryshortbottlenecksusingarepresentativewebapplicationbenchmark(RUBBoS).Besides,wevalidatetheaccuracyofoureventmScopeMonitorsanddemonstrateavailabilityandflexibilityofmilliScopethroughseveralevaluations.
Stark:OptimizingIn-MemoryComputingForDynamicDatasetCollections ShenLi(IBMResearch),MdTanvirAlAmin(UIUC),RaghuGanti(IBMResearch),MudhakarSrivatsa(IBMResearch),ShaohanHu(IBMResearch),YiranZhao(UIUC),TarekAbdelzaher(UIUC)
Emergingdistributedin-memorycomputingframeworks,suchasApacheSpark,canprocessahugeamountofcacheddatawithinseconds.Thisremarkablyhighefficiencyrequiresthesystemtowellbalancedataacrosstasksandensuredatalocality.However,itischallengingtosatisfytheserequirementsforapplicationsthatoperateonacollectionofdynamicallyloadedandevicteddatasets.Thedynamicsmayleadtotime-varyingdatavolumeanddistribution,whichwouldfrequentlyinvokeexpensivedatare-partitionandtransferoperations,resultinginhighoverheadandlargedelay.Toaddressthisproblem,wepresentStark,a
![Page 57: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/57.jpg)
57
systemspecificallydesignedforoptimizingin-memorycomputingondynamicdatasetcollections.Starkenforcesdatalocalityfortransformationsspanningmultipledatasets(e.g.,joinandcogroup)toavoidunnecessarydatareplicationsandshuffles.Moreover,toaccommodatefluctuatingdatavolumeandskeweddatadistribution,Starkdeliverselasticityintopartitionstobalancetaskexecutiontimeandreducejobmakespan.Finally,Starkachievesboundedfailurerecoverylatencybyoptimizingthedatacheckpointingstrategy.Evaluationsona50-serverclustershowthatStarkreducesthejobmakespanby4Xandimprovessystemthroughputby6XcomparedtoSpark.
CRESON:CallableandReplicatedSharedObjectsoverNoSQL PierreSutra(TélécomSudParis,CNRS,UniversitéParis-Saclay,France),EtienneRivière(UniversityofNeuchatel),CristianCotes(UniversitatRoviraiVirgili),MarcSánchezArtigas(UniversitatRoviraiVirgili),PedroGarciaLopez(UniversitatRoviraiVirgili),EmmanuelBernard(RedHat),WilliamBurns(RedHat),GalderZamarreno(RedHat)
TheabilitytoshareandpersistobjectssimplifiesthedesignofapplicationsinCloudenvironments.StoringobjectsonaNoSQLdatabaseensuresavailabilityandscalability.WhenObject-NoSQLMappingisperformedattheclientside,objectsthatareaccessedconcurrentlyarerepeatedlyconvertedbetweentheirin-memoryandserializedrepresentations.Thisnegativelyimpactsperformanceandincreasesreplicationcosts.WedescribeinthispaperthedesignofCRESON,asystemsupportingcallableobjectsoverNoSQL,inwhichapplicationobjectsaremappedandinstantiateddirectlyonthestoragenodes.CRESONsupportscompositionbyreferenceandensuresstrongconsistency.ObjectsarereplicatedandmaintainedcoherentusingStateMachineReplication.TheimplementationofCRESONleveragesthesupportofalistenablekey-valuestore(LKVS),anovelNoSQLstorageabstractionthatweintroduceinthispaper.WediscusstheperformanceandcomplexityofusingCRESONwiththeexampleoftheportageofapersonalcloudstorageservice,initiallydevelopedusingobject-relationalmappingoverashardedPostgreSQLdatabase.OurresultsshowthatCRESONoffersasimplerprogrammingexperiencebothintermsoflearningtimeandlinesofcode,whileperformingbetteronaverageandbeingmorescalable.
VirtualizedNetworkCodingFunctionsontheInternet LinquanZhang(UniversityofCalgary),ShangqiLai(TheUniversityofHongKong),ChuanWu(TheUniversityofHongKong),ZongpengLi(UniversityofCalgary),ChuanxiongGuo(MicrosoftResearch)
Networkcodingisafundamentaltoolthatenableshighernetworkcapacityandlowercomplexityinroutingalgorithms,byencouragingthemixingofinformationflowsinthemiddleofanetwork.ImplementingnetworkcodinginthecoreInternetissubjecttopracticalconcerns,sinceInternetroutersareoftenoverwhelmedbypacketforwardingtasks,leavinglittleprocessingcapacityforcodingoperations.Inspiredbytherecentparadigmofnetworkfunctionvirtualization,weproposeimplementingnetworkcodingasanewnetworkfunction,anddeployingsuchcodingfunctionsingeo-distributedclouddatacenters,topracticallyenablenetworkcodingontheInternet.Wetargetmulticastsessions(includingunicastflowsasspecialcases),strategicallydeployrelaynodes(networkcodingfunctions)inselecteddatacentersbetweensendersandreceivers,andembracehighbandwidthefficiencybroughtbynetworkcodingwithdynamiccodingfunctiondeployment.Wedesignandimplementthenetworkcodingfunctionontypicalvirtualmachines,featuringefficientpacketprocessing.Weproposeanefficientalgorithmforcodingfunctiondeployment,scalinginandout,inthepresenceofsystemdynamics.Real-worldimplementationonAmazonEC2andLinodedemonstratessignificantthroughputimprovementandhigherrobustnessofmulticastviacodingfunctionsaswellasefficiencyofthedynamicdeploymentandscalingalgorithm.
Research3:SecurityandPrivacyinDistributedSystemsIConsensusRobustnessandTransactionDe-AnonymizationintheRippleCurrencyExchangeSystem AdrianoDiLuzio(SapienzaUniversityofRome),AlessandroMei(SapienzaUniversityofRome),JulindaStefa(SapienzaUniversityofRome)
Distributedfinancialsystemsareradicallychangingthewaywedobusinessandspendourmoney.Ripple,inparticular,isuniqueinitskind.Itisbuiltonconsensusandtrustamongitspeersand,differentlyfrommanyothersystems,itallowsexchangingbothfiatcurrenciesandgoods.Itdoessobystoringtheaccountsofitsusers,theirbalances,andallthetransactionsinadistributedledger,publiclyaccessible.Inthispaperweperformforthefirsttimeanin-depthstudyoftheRippleexchangesystemanditspublicdistributedledger.Weanalyzepayments,thestructureofpaymentpaths,andtheroleofimportantpeersinthesystemsuchasGateways(theequivalentofBanks)andMarketMakersthatallowcross-currencyexchangeamongusers.TocompletethestudyoftheecosystemweanalyzetheinternalstreamofeventsinRippleandshowthatthewholesystemreliesonasurprisinglysmallnumberofactivevalidators,raisingseveralconcernsontherobustnessandontheactualfairnessofthesystem.Finally,weshowhowdistributedfinancialsystemscanjeopardizetheprivacyoftheirusers.Byexaminingthefirst3yearsofRipplehistory(morethan500GBofdata),weshowthatevenapproximateinformationonasinglepaymentcanuncover,withincredibleaccuracy,thewholefinanciallifeoftheuser.Forexample,thisallowsanyonewhooverhearsourorderforaLatteatourfavouritebartogaincompleteandunlimitedaccesstoourbalance,ourpreviousandfuturepayments,ourmonthlyincome,aswellastheplaceswhereweshopandthepeoplewetrust.
LearningprivacyhabitsofPDSowners BikashSingh(UniversityofInsubria),BarbaraCarminati(universityofinsubria),ElenaFerrari(universityofinsubria)
TheconceptofPersonalDataStorage(PDS)hasrecentlyemergedasanalternativeandinnovativewayofmanagingpersonaldataw.r.t.theservice-centriconecommonlyusedtoday.ThePDSoffersauniquelogicalrepository,allowingindividualstocollect,store,andgiveaccesstotheirdatatothirdparties.TheresearchonPDShassofarmainlyfocusedontheenforcementmechanisms,thatis,onhowuserprivacypreferencescanbeenforced.Incontrast,thefundamentalissueofpreferencespecificationhasbeensofarnotdeeplyinvestigated.Inthispaper,wedoastepinthisdirectionbyproposingdifferentlearningalgorithmsthatallowafine-grainedlearningoftheprivacyaptitudesofPDSowners.Thelearnedmodelsarethenusedtoanswerthirdpartyaccessrequests.Theextensiveuserstudieswehaveperformedshowtheeffectivenessoftheproposedapproach.
City-Hunter:HuntingSmartphonesinUrbanAreas XuefengLiu(HongKongPolytechnicUniversity),JiaqiWen(HongKongPolytechnicUniversity),ShaojieTang(UniversityofTexasatDallas),JiannongCao(HongKongPolytechnicUniversity),JiaxingShen(HongKongPolytechnicUniversity)
ThesecurityissueofpublicWiFiisgainingmoreandmoreconcern.Bylisteningtoproberequests,anadversarycanobtaintheSSIDlistoftheAPstowhichasmartphonepreviouslyconnected,andutilizesthisinformationtotrickthesmartphoneintoassociatingtoit.However,withtheenhancementofsecuritylevel,mostsmartphonesnowdonotproactivelydisclosetheirSSIDlists,makingtheseattacksobsolete.Inthispaper,weproposeCity-Hunter,anattackerthatcanlurenearbysmartphoneswithoutknowingtheirSSIDinformation.City-HunterestablishesandmaintainsanSSIDdatabasebyintegratingbothofflineandonlineinformation.Meanwhile,itsmartlychoosessomeSSIDstohitasmartphoneaccordingtothepastrecordandfreshness.WeevaluatetheperformanceofCity-
![Page 58: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/58.jpg)
58
Hunterindifferentpublicplaces.TheresultsdemonstratethatCity-Hunterisabletosuccessfullyhit12%_18%smartphoneswithoutknowingtheirSSIDinformation,whichisabout4_8timesimprovementcomparedtothesimilarattackslikeKARMAandMANA.
WhenSeeingIsn'tBelieving:OnFeasibilityandDetectabilityofScapegoatinginNetworkTomography ShangqingZhao(UniversityofSouthFlorida),ZhuoLu(UniversityofSouthFlorida),CliffWang(NorthCarolinaStateUniversity/ArmyResearchOffice)
Networktomographyisavitaltooltoestimatelinkqualitiesfromend-to-endnetworkmeasurements.Animplicitassumptioninnetworktomographyisthatobservedmeasurementsindeedreflecttheaggregateoflinkperformance(i.e.,seeingisbelieving).However,itisnotguaranteedtodaythatthereexistsnoanomaly(e.g.,maliciousautonomoussystemsandinsiderthreats)inlarge-scalenetworks.Maliciousnodescanintentionallymanipulatelinkmetricsviadelayingordroppingpacketstoaffectmeasurements.Willsuchanassumptionrenderavulnerabilitywhenfacingattackers?Theproblemisofessentialimportanceinthatnetworktomographyisdevelopedtowardseffectivenetworkdiagnosticsandfailurerecovery.Inthispaper,wedemonstratethatthevulnerabilityisrealandproposeanewattackstrategy,calledscapegoating,inwhichmaliciousnodescansubstantiallydamageanetwork(e.g.,delayingpackets)andatthesametimemaliciouslymanipulateend-to-endmeasurementresultssuchthatalegitimatenodeismisleadinglyidentifiedastherootcauseofthedamage(therebybecomingascapegoat)undernetworktomography.Weformulatethreebasicscapegoatingapproachesandshowunderwhatconditionsattackscanbesuccessful.Wealsorevealconditionstodetectsuchattacks.Ourtheoreticalandexperimentalresultsshowthatsimplytrustingmeasurementsleadstoscapegoatingvulnerabilities.Thus,existingmethodsshouldberevisitedaccordinglyforsecurityinvariousapplications.
YouCanHearButYouCannotSteal:DefendingagainstVoiceImpersonationAttacksonSmartphones SiChen(UniversityatBuffalo/WestChesterUniversity),KuiRen(UniversityatBuffalo),SixuPiao(UniversityatBuffalo),CongWang(CityUniversityofHongKong),QianWang(WuhanUniversity),JianWeng(JinanUniversity),LuSu(UniveristyatBuffalo),AzizMohaisen(UniversityatBuffalo)
Voice,asaconvenientandefficientwayofinformationdelivery,hasasignificantadvantageovertheconventionalkeyboard-basedinputmethods,especiallyonsmallmobiledevicessuchassmartphonesandsmartwatches.However,thehumanvoicecouldoftenbeexposedtothepublic,whichallowsanattackertoquicklycollectsoundsamplesoftargetedvictimsandfurtherlaunchvoiceimpersonationattackstospoofthosevoicebasedapplications.Inthispaper,weproposethedesignandimplementationofarobustsoftware-onlyvoiceimpersonationdefensesystem,whichistailoredformobileplatformsandcanbeeasilyintegratedwithexistingoff-the-shelfsmartdevices.Inoursystem,weexploremagneticfieldemittedfromloudspeakersastheessentialcharacteristicfordetectingmachine-basedvoiceimpersonationattacks.Furthermore,weuseastate-of-the-artautomaticspeakerverificationsystemtodefendagainsthumanimitationattacks.Ouradvancedsensorydataprocessingtechniqueachievesfastauthenticationspeed(6.1s)whichmakesitsuitableformobileplatforms.Finally,ourevaluationresultsshowthatoursystemachievessignificantlyhighaccuracy(100%)andlow(0%)equalerrorrates(EERs)indetectingthemachinebasedvoiceimpersonationattackonsmartphones.
FlowReconnaissanceviaTimingAttacksonSDNSwitches ShengLiu(UniversityofNorthCarolinaatChapelHill),MichaelReiter(UniversityofNorthCarolinaatChapelHill),VyasSekar(CarnegieMellonUniversity)
Whenencounteringapacketflowforwhichithasnocoveringrule,asoftware-definednetworking(SDN)switchrequestsanappropriaterulefromitscontroller;thisrequestdelaystheroutingoftheflowuntilthecontrollerresponds.Weshowthatthisdelaygivesrisetoatimingsidechannelinwhichanattackercantestfortherecentoccurrenceofatargetflowbyjudiciouslyprobingtheswitchwithforgedflowsandusingthedelaystheysuffertodiscernwhethercoveringruleswerepreviouslyinstalledintheswitch.WedevelopaMarkovmodelofanSDNswitchtopermittheattackertoselectthebestprobe(orprobes)toinferwhetheratargetflowhasrecentlyoccurred.Ourmodelcapturescomplexitiesrelatedtoruleevictionstomakeroomforotherrules;ruletimeoutsduetoinactivity;thepresenceofmultiplerulesthatapplytooverlappingsetsofflows;andrulepriorities.Weshowthatourmodelpermitsdetectionoftargetflowswithconsiderableaccuracyinmanycases.
Research4:CloudComputingandDataCenterSystemsAStudyofLong-TailLatencyinn-TierSystems:RPCvs.AsynchronousInvocations QingyangWang(LouisianaStateUniversity),Chien-AnLai(GeorgiaTech),YasuhikoKanemasa(FujitsuLaboratoriesLtd.),ShungengZhang(LouisianaStateUniversity),CaltonPu(GeorgiaTech)
Long-taillatencyofweb-facingapplicationscontinuestobeaseriousproblem.Mostofthepreviouslypublishedresearchaddressestwoclassesoflonglatencyproblems:unevenworkloadssuchaswebsearch,andresourcesaturationinsinglenodes.Wedescribeanexperimentalstudyofathirdclassoflongtaillatencyproblemsthatarespecifictodistributedsystems:Cross-TierQueueOverflow(CTQO)duetoacombinationofmillibottlenecks(withsub-secondduration)andtightly-coupledserversinn-tiersystems(e.g.,Apache,Tomcat,andMySQL)usingRPC-stylerequest-responsecommunications.Ourexperimentsshowthattheappearanceofmillibottlenecks(e.g.,createdbyshortworkloadbursts)inoneserveroftencausesanotherserver(whichhasnosaturatedresources)inthesynchronousinvocationchaintofillupitsqueues(CTQO)anddroppackets,creatingverylongresponsetimequeries.CTQOcanbereducedoravoidedbyreplacingtheserverdroppingpacketswithanasynchronousserver.Insynchronousn-tiersystemexperiments,longtaillatencyduetoCTQOcanbereproducedconsistentlyatutilizationaslowas43%.Incontrast,whenalln-tierserversarereplacedbyasynchronousversions,CTQOandconsequentdroppedpacketsremainabsentatutilizationlevelsashighas83%,despitethesamemillibottlenecks.
RainorShine?-MakingSenseofCloudyReliabilityData IyswaryaNarayanan(ThePennsylvaniaStateUniversity),BikashSharma(Microsoft),DiWang(Microsoft),SriramGovindan(Microsoft),LauraCaulfield(Microsoft),AnandSivasubramaniam(ThePennsylvaniaStateUniversity),AmanKansal(Microsoft),JieLiu(Microsoft),BadriddineKhessib(Microsoft),KushagraVaid(Microsoft)
Clouddatacentersmustensurehighavailabilityforthehostedapplicationsandfailurescanbethebaneofdatacenteroperators.Understandingthewhat,whenandwhyoffailurescanhelptremendouslytomitigatetheiroccurrenceandimpact.Failurescan,however,dependonnumerousspatialandtemporalfactorsspanninghardware,workloads,supportfacilities,andeventheenvironment.Onehastorelyonfailuredatafromthefieldtoquantifytheinfluenceofthesefactorsonfailures.Towardsthisgoal,wecollectfailuresdataalongwithmanyparametersthatmightinfluencefailuresfromtwolargeproductiondatacenterswithvery
![Page 59: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/59.jpg)
59
diversecharacteristics.Weshowthatmultiplefactorssimultaneouslyaffectfailures,andthesefactorsmayinteractinnon-trivialways.Thismakesconventionalapproachesthatstudyaggregatecharacteristicsorsingleparameterinfluences,ratherinaccurate.Instead,webuildamulti-factoranalysisframeworktosystematicallyidentifyinfluencingfactors,quantifytheirrelativeimpact,andhelpinmoreaccuratedecisionmakingforfailuremitigation.Wedemonstratethisapproachforthreeimportantdecisions:sparecapacityprovisioning,comparingthereliabilityofhardwareforvendorselection,andquantifyingflexibilityindatacenterclimatecontrolforcost-reliabilitytrade-offs.
Right-sizingGeo-distributedDataCentersforAvailabilityandLatency IyswaryaNarayanan(ThePennsylvaniaStateUniversity),AmanKansal(Microsoft),AnandSivasubramaniam(ThePennsylvaniaStateUniversity)
Weshowclouddevelopershowtorightsizedatacenter(DC)capacityforgeo-distributedapplicationsdeployedonseveralmulti-megawattDCs,possiblyalsousingmanysmalleredgeDCs.Notethatcapacityconsiderationsforageo-distributedinfrastructuredonotdecomposeintoindividualDCcapacityplanning.WhenedgeDCsareused,heterogeneousavailabilityandcostsaffectthecapacitysplitbetweentheedgeandcoreDCs.Non-uniformspatialdistributionofclientsandinterdependencebetweenlatencyandavailabilityconstraintsmakeitnon-trivialtoprovisiontherightcapacityateachDC.Wedevelopageo-distributedcapacityplanningframeworktocapturethekeyfactorsthatinfluencecapacity,rangingfromapplicationdemandpatterns,latencyandavailabilityrequirements,DCcost-availabilitytrade-offs,anddatareplicationoverheads.WeapplyourframeworktoarealisticapplicationandDCinfrastructuresettingtogatherinsightsintohowcapacityshouldbeprovisionedandallocatedacrossDCsforarepresentativesetofrequirementsandcosts.
PerformanceDrivenResourceSharingMarketsfortheSmallCloud Sung-HanLin(UniversityofSouthernCalifornia),RanjanPal(UniversityofSouthernCalifornia),MarcoPaolieri(UniversityofSouthernCalifornia),LeanaGolubchik(UniversityofSouthernCalifornia)
Small-scaleclouds(SCs)oftensufferfromresourceunder-provisioningduringpeakdemand,leadingtoinabilitytosatisfyservicelevelagreements(SLAs)andconsequentlossofcustomers.OneapproachtoaddressthisproblemisforasetofautonomousSCstoshareresourcesamongthemselvesinacostinducedcooperativefashion,therebyincreasingtheirindividualcapacities(whenneeded)withouthavingtosignificantlyinvestinmoreresources.Acentralproblem(inthiscontext)ishowtoproperlyshareresources(foraprice)toachieveprofitableservicewhilemaintainingcustomerSLAs.Toaddressthisproblem,inthispaper,weproposetheSC-Shareframeworkthatutilizestwointeractingmodels:(i)astochasticperformancemodelthatestimatestheachievedperformancecharacteristicsundergivenSLArequirements,and(ii)amarket-basedgame-theoreticmodelthat(asshownempirically)convergestoefficientresourcesharingdecisionsatmarketequilibrium.Ourresultsincludeextensiveevaluationsthatillustratetheutilityoftheproposedframework.
Fault-scalableVirtualizedInfrastructureManagement MukilKesavan(VMwareInc.),AdaGavrilovska(GeorgiaInstituteofTechnology),KarstenSchwan(GeorgiaInstituteofTechnology)
Large-scalevirtualizeddatacentersrequireconsiderableautomationininfrastructuremanagementinordertooperateefficiently.Automationisimpaired,however,bythefactthatdeploymentsarepronetomultipletypesofsubtlefaultsduetohardwarefailures,softwarebugs,misconfiguration,crashes,performancedegradedhardware,etc.ExistingInfrastructure-as-a-Service(IaaS)managementstacksincorporatelittletonoresiliencemeasurestoshieldendusersfromsuchcloudproviderlevelfailuresandpoorperformance.ThispaperproposesandevaluatesextensionstoIaaSstacksthatmaskfaultsinafault-agnosticmannerwhileensuringthattheoverheadscanbeproportionaltoobservedfailurerates.Wealsodemonstratethatinfrastructureautomationservicesandend-userapplicationscanuseservice-specificknowledge,togetherwithournewinterface,toachievebetteroutcomes.
DeltaCFS:BoostingDeltaSyncforCloudStorageServicesbyLearningfromNFS QuanluZhang(PekingUniversity),ZhenhuaLi(TsinghuaUniversity),ZhiYang(PekingUniversity),ShenglongLi(PekingUniversity),YangzeGuo(PekingUniversity),YafeiDai(PekingUniversity),ShouyangLi(PekingUniversity)
Cloudstorageservices,suchasDropbox,iCloudDrive,GoogleDrive,andMicrosoftOneDrive,havegreatlyfacilitateduserssynchronizingfilesacrossheterogeneousdevices.Amongthem,Dropbox-likeservicesareparticularlybeneficialowingtothedeltasyncfunctionalitythatstrivestowardsgreaternetwork-levelefficiency.However,whendeltasynctradescomputationoverheadfornetwork-trafficsaving,thetradeoffcouldbehighlyunfavorableundersometypicalworkloads.Werefertothisproblemastheabuseofdeltasync.Toaddressthisproblem,weproposeDeltaCFS,anovelfilesyncframeworkforcloudstorageservicesbylearningfromthedesignofconventionalNFS(NetworkFileSystem).Specifically,wecombinedeltasyncwithNFS-likefileRPCinanadaptivemanner,thussignificantlycuttingcomputationoverheadonboththeclientandserversideswhilepreservingthenetwork-levelefficiency.DeltaCFSalsoenablesaneatdesignforguaranteeingcausalconsistencyandfine-grainedversioncontroloffiles.InourFUSE-basedprototypesystem(whichisopen-source),DeltaCFSoutperformsDropboxbygeneratingupto11_lessdatatransferandupto100_lesscomputationoverheadunderconcernedworkloads.
Research5:EdgeandFogComputingCachier:Edge-cachingforrecognitionapplications UtsavDrolia(CarnegieMellonUniversity),KatherineGuo(BellLabs),JiaqiTan(Nokia),RajeevGandhi(CarnegieMellonUniversity),PriyaNarasimhan(CarnegieMellonUniversity)
Recognitionandperception-basedmobileapplications,suchasimagerecognition,areontherise.Theseapplicationsrecognizetheuserssurroundingsandaugmentitwithinformationand/ormedia.Theseapplicationsarelatencysensitive.Theyhaveasoft-realtimenature-lateresultsarepotentiallymeaningless.Ontheonehand,giventhecomputeintensivenatureofthetasksperformedbysuchapplications,executionistypicallyoffloadedtothecloud.Ontheotherhand,offloadingsuchapplicationstothecloudincursnetworklatency,whichcanincreasetheuser-perceivedlatency.Consequently,edge-computinghasbeenproposedtoletdevicesoffloadintensivetaskstoedge-serversinsteadofthecloud,toreducelatency.Inthispaper,weproposeadifferentmodelforusingedge-servers.Weproposetousetheedgeasaspecializedcacheforrecognitionapplicationsandformulatetheexpectedlatencyforsuchacache.Weshowthatusinganedge-serverlikeatypicalweb-cache,forrecognitionapplications,canleadtohigherlatencies.WeproposeCachier,asystemthatusesthecachingmodelalongwithnoveloptimizationstominimizelatencybyadaptivelybalancingloadbetweentheedgeandthecloudbyleveragingspatiotemporallocalityofrequests,usingofflineanalysisofapplications,andonlineestimatesofnetworkconditions.WeevaluateCachierforimage-recognitionapplicationsandshowthatourtechniquesyield3xspeed-upinresponsiveness,andperformaccuratelyoverarangeofoperatingconditions.Tothebestofourknowledge,thisisthefirstworkthatmodelsedge-serversascachesforcomputeintensiverecognitionapplications,andCachieristhefirstsystemthatusesthismodeltominimizelatencyfortheseapplications.
![Page 60: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/60.jpg)
60
ContentCentricPeerDataSharinginPervasiveEdgeComputingEnvironments XintongSong(PekingUniversity),YaodongHuang(StonyBrookUniversity),QianZhou(StonyBrookUniversity),FanYe(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity),XiaomingLi(PekingUniversity)
Theproliferationanddailycongregationofmodernmobiledeviceshavecreatedabundantopportunitiesforpeeredgedevicestosharevaluabledatawitheachother.Theshortcontactdurations,relativelysmallsharingsizes,anduncertaindataavailability,demandagile,lightweightpeerbaseddatasharing.Inthispaper,weproposePeerDataSharing(PDS)thatenablesedgedevicestodiscoverwhichdataexistinnearbypeers,andretrieveinteresteddatarobustlyandefficiently.PDSusesnovellingeringqueries,mixedcastanden-routemessagerewritingtechniquestominimizeredundanttransmissionsandmaximizeopportunisticoverhearingthuscachingindatadiscoveryandretrieval.ExtensiveevaluationsbasedonanAndroidprototypeshowthatPDSdiscoversandretrievesalmost100%dataintensofseconds,andremainsrobustdespitewirelesscontention,simultaneousconsumerrequestsandusermobility.
FLARE:CoordinatedRateAdaptationforHTTPAdaptiveStreaminginCellularNetworks YoungbinIm(UniversityofColoradoatBoulder),JinyoungHan(HanyangUniversity),JiHoonLee(JuniKorea),YoonKwon(Kakao),CarleeJoe-Wong(CarnegieMellonUniversity),TaekyoungKwon(SeoulNationalUniversity),SangtaeHa(UniversityofColoradoatBoulder)
Fogcomputingisanemergingarchitecturethataimstorunapplicationsonmultipledevicesthatlieonacontinuumfromcentralizedcloudserverstopersonaluserdevices.Thesearchitecturesallowapplicationstooptimizeovertheinformationstoredateachtypeofdeviceanddividetheirfunctionalitiesbasedonthedevicecapabilities.Wedemonstratethebenefitsofthisapproachformobilevideostreaming.ExistingHAS(HTTPadaptivestreaming)techniques,usedbypopularvideoserviceproviders,oftensufferfromproblemslikeunstablevideoqualityandsuboptimalresourceutilization.Wefindthatalackofcoordinationpreventsbothclient-andnetwork-sideHAStechniquesfromsolvingthem.However,ourfogapproachcanexploitexistingtelecommunicationAPIs,whichexposenetworkcapabilitiestoapplications,inordertocoordinatebetweentheclientandnetwork.OurcoordinatedHASsolution,FLARE,optimizesthetotalutilityofallclientsinacellwhilemaintainingstablevideoqualityandsupportinguser-anddevice-specificneeds.WeimplementFLAREonacommodityLTEfemtocellandusetheimplementationtoconductthefirstcomparisonofHASplayersonanLTEfemtocell.Byconductingextensiveexperimentsusingthens-3simulator,wealsodemonstratethatFLARE(i)enhancestheaveragevideobitrate,(ii)achievesstablevideoquality,and(iii)balancesthethroughputofsimultaneousvideoanddataflows,comparedtootherrepresentativeHASsolutions.
NetworkedDroneCamerasforSportsStreaming XiaoliWang(PrincetonUniversity),AakankshaChowdhery(PrincetonUniversity),MungChiang(PrincetonUniversity)
Anetworkofdronecamerascanbedeployedtocoverliveevents,suchashigh-actionsportsgameplayedonalargefield,butmanagingnetworkeddronecamerasinreal-timeischallenging.Distributedapproachesyieldsuboptimalsolutionsfromlackofcoordinationbutcoordinationwithacentralizedcontrollerincursround-triplatenciesofseveralhundredsofmillisecondsoverawirelesschannel.Weproposeafog-networkingbasedsystemarchitecturetoautomaticallycoordinateanetworkofdronesequippedwithcamerastocaptureandbroadcastthedynamicallychangingscenesofinterestinasportsgame.Wedesignbothoptimalandpracticalalgorithmstobalancethetradeoffbetweentwometrics:coverageofthemostimportantscenesandstreamedvideobitrate.Tocompensatefornetworkround-triplatencies,thecentralizedcontrollerusesapredictiveapproachtopredictwhichlocationsthedronesshouldcovernext.Thecontrollermaximizesvideobitratebyassociatingeachdronetoanoptimallymatchedserveranddynamicallyre-assignsdronesasrelaynodestoboostthethroughputinlow-throughputscenarios.Thisdynamicassignmentatcentralizedcontrolleroccursatslowertime-scalepermittedbyround-triplatencies,whilethepredictiveapproachanddroneslocaldecisionensuresthatthesystemworksinreal-time.Experimentalresultsovertensofflightsonthefieldsuggestoursystemcanachievereallygoodperformance,forexample,8dronescanachieveatradeoffof94%coverageand(onaverage)2Kvideosupportat20Mbpsbyoptimizingbetweencoverageandthroughput.Bydynamicallyallocatingdronestocoverthegameoractasrelays,oursystemalsodemonstratesa2xgainoversystemsmaximizingstaticcoveragealonethatachievesonly9Mbpsvideothroughput.
Chronus:ConsistentDataPlaneUpdatesinTimedSDNs JiaqiZheng(NanjingUniversity),GuihaiChen(NanjingUniversity),StefanSchmid(AalborgUniversity),HaipengDai(NanjingUniversity),JieWu(TempleUniversity)
Software-DefinedNetworks(SDNs)introduceinterestingnewopportunitiesinhownetworkroutescanbedefined,verified,andchangedovertime.Yetdespitethelogicallycentralizedperspectiveoffered,anSDNstillneedstobeconsideredadistributedsystem:ruleupdatescommunicatedfromthecontrollertotheindividualswitchestraverseanasynchronousnetworkandmayarriveout-of-order,andhenceleadto(temporaryorpermanent)inconsistencies.Accordingly,theconsistentnetworkupdateproblemhasrecentlyreceivedmuchattention.MotivatedbytheadventoftightlysynchronizedSDNs,weinthispaperinitiatethestudyofalgorithmsforconsistentnetworkupdatesintimedSDNsSDNsinwhichindividualnodeupdatescanbescheduledatspecifictimes.ThispaperpresentsChronus,whichisbasedonprovablycongestion-andloop-freeupdateschedulingalgorithms,andavoidstheflowtablespaceheadroomrequiredbyexistingtwophaseupdateapproaches.WeformulatetheMinimumUpdateTimeProblem(MUTP)asanoptimizationprogram.Weproposeatreealgorithmtocheckthefeasibilityandagreedyalgorithmtofindaupdatesequenceinpolynomialtime.ExtensiveexperimentsonMininetandnumericalsimulationsshowthatChronuscansubstantiallyreducetransientcongestionby75%andsaveover60%oftherulescomparedtothestateoftheart.
DistributedDeepNeuralNetworksovertheCloud,theEdgeandEndDevicesSuratTeerapittayanon(HarvardUniversity),BradleyMcDanel(HarvardUniversity),H.T.Kung(HarvardUniversity)
Weproposedistributeddeepneuralnetworks(DDNNs)overdistributedcomputinghierarchies,consistingofthecloud,theedge(fog)andenddevices.Whilebeingabletoaccommodateinferenceofadeepneuralnetwork(DNN)inthecloud,aDDNNalsoallowsfastandlocalizedinferenceusingshallowportionsoftheneuralnetworkattheedgeandenddevices.Moreover,viadistributedcomputing,DDNNsenhancedataprivacyandsystemfaulttoleranceforDNNapplications.Whensupportedbyascalabledistributedcomputinghierarchy,aDDNNcanscaleupinneuralnetworksizeandscaleoutingeographicalspan.InimplementingaDDNN,wemapsectionsofaDNNontoadistributedcomputinghierarchy.Byjointlytrainingthesesections,weminimizecommunicationandresourceusagefordevicesandmaximizeusefulnessofextractedfeatureswhichareutilizedinthecloud.Asaproofofconcept,weshowaDDNNcanexploitgeographicaldiversityofsensorstoimproverecognitionaccuracyandreducecommunicationcost.Inourexperiment,comparedwiththetraditionalmethodofoffloadingrawsensordatatobeprocessedinthecloud,DDNNlocallyprocessesmostsensordataonenddevicesandisabletoreducethecommunicationcostbyafactorofover20x.
Research6:DistributedGreenComputingandEnergyManagement
![Page 61: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/61.jpg)
61
DynamicControlofFlowCompletionTimeForPowerEfficientDataCenterNetworks KuangyuZheng(TheOhioStateUniversity),XiaoruiWang(TheOhioStateUniversity)
Datacenternetwork(DCN)canconsumeasignificantamountofpower(e.g.,10%to20%)inlarge-scaledatacenters.ToreducethepowerconsumptionofDCN,trafficconsolidationhasbeenrecentlyproposedasaneffectiveapproachtoreducethenumberofDCNdevicesinuse.However,existingconsolidationapproachesdonotsufficientlyconsidertheflowcompletiontime(FCT)requirement.Ononehand,missingtheFCTdeadlinescancauseseriousviolationofservice-levelagreement,especiallyfordelay-sensitivenetworkingservices,suchaswebsearchandE-commerce.Ontheotherhand,keepingallthedevicesontomakeFCTsmuchshorterthanthedesiredrequirementsisunnecessarybecause1)usersmaynotbeabletoperceivethedifference,and2)suchagreedystrategycanleadtounnecessarilyhighDCNpowerconsumptionandthusmoreelectricitycosts.Inthispaper,weproposeFCTcon,adynamicFCTcontrolstrategyforDCNpoweroptimization.FCTconisdesignedrigorouslybasedoncontroltheorytodynamicallycontroltheFCTofdelay-sensitivetrafficflowsexactlytorequirements,suchthatthedesiredFCTperformanceisguaranteedwhilethemaximumamountofDCNpowersavingscanbeachieved.Resultsfrombothhardwareexperimentsandsimulationevaluationsdemonstratethat,comparedtothestate-of-the-artDCNpoweroptimizationschemes,FCTconcanimprovetheDCNFCTperformance,whileachievingnearlythesameorevenmorepowersavings.Consequently,FCTconcanresultinmorethan22.0%to62.2%extranetprofitsforadatacenterwith50Kservers.
OnEnergy-EfficientCongestionControlforMultipathTCP JiaZhao(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity),HaiyangWang(UniversityofMinnesotaDuluth)
Multipathtransportprotocols,e.g.MultipathTCP(MPTCP),enabletransmissionviamultipleroutesbetweenanend-to-endconnectiontoimproveresourceusageofregularTCP.Duetotheincreasingconcerningreencomputing,therehasbeensignificantinterestindesigningenergy-efficientmultipathtransport.ForexistingMPTCPcongestioncontrolalgorithms,theresearchcommunitystilllacksacomprehensiveunderstandingofwhichcomponentsinsuchanalgorithmplaythefundamentalroleinenergyefficiency,howvariousalgorithmscompareagainsteachotherfromenergy-consumingperspective,orwhetherthereexistpotentiallybettersolutionsforenergysaving.Inthispaper,wetakeafirststeptoanswerthesequestions.BasedontheMPTCPLinuxkernelexperiments,wefirstsummarizethattheenergyconsumptionisrelatedtothreeaspects:averagethroughput,pathdelayanddifferentnetworkscenarios.Inordertobridgecongestioncontroltothethreeaspects,weanalyzetheexistingalgorithmsandcapturetheessentialparametersofmultipathcongestioncontrolmodelrelatedtoMPTCPsenergyefficiency.Thenwedesignawindowincreasefactortoshifttraffictolow-delayenergy-efficientpaths.Wefurtherextendthisdesignbyusinganenergy-awarecompensativeparametertofitthegeneralhierarchicalInternettopology.Weevaluatetheperformanceofexistingmultipathcongestioncontrolalgorithmsandourproposedalgorithmindifferentnetworkscenarios.Theresultsshowenergyefficiencyofourdesign.
AMechanismforCooperativeDemand-SideManagement GuangchaoYuan(Microsoft),Chung-WeiHang(IBM),MichaelHuhns(UniversityofSouthCarolina),MunindarSingh(NorthCarolinaStateUniversity)
Demand-sidemanagement(DSM)isanimportantthemeintheSmartGridandoffersthepossibilityoflevelingpowerconsumptionwithitsattendantbenefitsofreducingcapitalexpenses.Thispaperdevelopsanalgorithmicmechanismthatreducespeaktotalconsumptionandencouragesprosocialbehavior,suchasexpressingflexibilityinonespowerconsumptionandreportingpreferencestruthfully.Ourobjectiveistoprovideatractable,budget-balancedmechanismthatpromotestruthtellingfromhouseholds.Theresultingmechanismistheoreticallyandempiricallyproventobeexantebudget-balanced,weaklyPareto-efficient,andweaklyBayesianincentive-compatible.Asimulationstudyverifiesthatthemechanismcouldlargelyreducethecomputationalcomplexitythattheoptimalallocationrequires,whilemaintainingapproximatelythesameperformance.Auserstudywith20subjectsfurthershowstheeffectivenessofthemechanisminpreventingparticipantsfromdefectingandincentivizingthemtorevealflexiblepreferences.
AHierarchicalFrameworkofCloudResourceAllocationandPowerManagementUsingDeepReinforcementLearning NingLiu(SyracuseUniversity),ZheLi(SyracuseUniversity),ZhiyuanXu(SyracuseUniversity),JielongXu(SyracuseUniversity),ShengLin(SyracuseUniversity),QinruQiu(SyracuseUniversity),JianTang(SyracuseUniversity),YanzhiWang(SyracuseUniversity)
Automaticdecision-makingapproaches,suchasreinforcementlearning(RL),havebeenappliedto(partially)solvetheresourceallocationproblemadaptivelyinthecloudcomputingsystem.However,acompletecloudresourceallocationframeworkexhibitshighdimensionsinstateandactionspaces,whichprohibittheusefulnessoftraditionalRLtechniques.Inaddition,highpowerconsumptionhasbecomeoneofthecriticalconcernsindesignandcontrolofcloudcomputingsystems,whichdegradessystemreliabilityandincreasescoolingcost.Aneffectivedynamicpowermanagement(DPM)policyshouldminimizepowerconsumptionwhilemaintainingperformancedegradationwithinanacceptablelevel.Thus,ajointvirtualmachine(VM)resourceallocationandpowermanagementframeworkarecriticaltotheoverallcloudcomputingsystem.Moreover,novelsolutionframeworkisnecessarytoaddresstheevenhigherdimensionsinstateandactionspaces.Inthispaper,weproposeanovelhierarchicalframeworkforsolvingtheoverallresourceallocationandpowermanagementproblemincloudcomputingsystems.TheproposedhierarchicalframeworkcomprisesaglobaltierforVMresourceallocationtotheserversandalocaltierfordistributedpowermanagementoflocalservers.Theemergingdeepreinforcementlearning(DRL)technique,whichcandealwithcomplicatedcontrolproblemswithlargestatespace,isadoptedtosolvetheglobaltierproblem.Furthermore,anautoencoderandanovelweightsharingstructureareadoptedtohandlethehigh-dimensionalstatespaceandacceleratetheconvergencespeed.Ontheotherhand,thelocaltierofdistributedserverpowermanagementscomprisesanLSTMbasedworkloadpredictorandamodel-freeRLbasedpowermanager,operatinginadistributedmanner.ExperimentresultsusingactualGoogleclustertracesshowthatourproposedhierarchicalframeworksignificantlysavesthepowerconsumptionandenergyusagethanthebaselinewhileachievingnoseverelatencydegradation.Meanwhile,theproposedframeworkcanachievethebesttrade-offbetweenlatencyandpower/energyconsumptioninaservercluster.
SunChase:Energy-EfficientRoutePlanningforSolar-PoweredEVs. LanduJiang(McGillUniversity),YuHua(HuazhongUniversityofScienceandTechnology),ChenMa(McGillUniversity),XueLiu(McGillUniversity)
Electricvehicles(EVs)playasignificantroleinthecurrenttransportationsystems.ThemainfactorthataffectsacceptanceofexistingEVmodelsistherangeanxietyproblemcausedbylimitedchargingstationsandlongrechargetimes.Recently,thesolar-poweredEVhasdrawnmanyattentionsduetobeingfreeofcharginglimitations.However,thesolarpoweredEVsmaystillstrugglewiththelimitedusebecauseofunpredictablesolaravailability.Forexample,shadingscausedbybuildingsandtreesalsopossiblydecreasethesolarpanelcellefficiency.Toaddressthis,weproposearouteplanningmethodforsolar-poweredEVstobalancetheenergyharvestingandconsumptionsubjecttotimeconstraint.Theideabehindoursolutionistoofferpower-awareoptimalrouting,whichmaximizestheon-roadenergyinputgivensolaravailabilityoneachroadsegment.Wefirstbuildasolaraccessestimationmodelusing3Dgeographicdataandthenemployamulti-criteria
![Page 62: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/62.jpg)
62
searchmethodtogenerateasetofParetocandidateroutes.Inordertoreducethesizeoftheset,weleveragethebisectkmeansclusteringalgorithmtoextractthemostrepresentativeParetorouteswithbettersolaravailability.Intheevaluation,wedevelopedavalidationplatformonthevehicleandleveragedmobilesensingtechniquestoexamineourproposedmodelinrealroadenvironments.Weconductedsimulationstoevaluateourproposedrouteplanningalgorithmusingreallifescenarios.Experimentalresultsdemonstratethatoursolarinputmodelisrobusttorealroadscenarios,andtheroutingalgorithmhasgreatpotentialtoprovideefficientservicesforsolar-poweredEVinthefuture.
Research7:InternetofThings,SmartCities,andCyber-PhysicalSystemsPersistentTrafficMeasurementThroughVehicle-to-InfrastructureCommunications HeHuang(SoochowUniversity),Yu-ESun(SoochowUniversity),ShigangChen(UniversityofFlorida),HongliXu(UniversityofScienceandTechnologyofChina),YianZhou(Google)
Measuringpointtrafficvolumeandpoint-to-pointtrafficvolumeinaroadsystemhasimportantapplicationsintransportationengineering.Theconnectedvehicletechnologiesintegratewirelesscommunicationsandcomputersintotransportationsystems,allowingwirelessdataexchangesbetweenvehiclesandroad-sideequipment,andenablinglarge-scale,sophisticatedtrafficmeasurement.Thispaperinvestigatestheproblemsofpersistentpointtrafficmeasurementandpersistentpoint-to-pointtrafficmeasurement,whichwerenotadequatelystudiedinthepriorart,particularlyinthecontextofintelligentvehicularnetworks.Weproposetwonovelestimatorsforprivacypreservingpersistenttrafficmeasurement:oneforpointtrafficandtheotherforpoint-to-pointtraffic.Theestimatorsaremathematicallyderivedfromthejoinresultoftrafficrecords,whichareproducedbytheelectronicroadsideunitswithprivacypreservingdatastructures.Weevaluateourestimationmethodsusingsimulationsbasedonbothrealtransportationtrafficdataandsyntheticdata.Thenumericalresultsdemonstratetheeffectivenessoftheproposedmethodsinproducinghighmeasurementaccuracyandallowingaccuracy-privacytradeoffthroughparametersetting.
TagBreathe:MonitorBreathingwithCommodityRFIDSystems YuxiaoHou(TheHongKongPolytechnicUniversity),YanwenWang(TheHongKongPolytechnicUniversity),YuanqingZheng(TheHongKongPolytechnicUniversity)
Breathmonitoringhelpsassessthegeneralpersonalhealthandgivescluestochronicdiseases.Yetcurrentbreathmonitoringtechnologiesareinconvenientandintrusive.Forinstance,typicalbreathmonitoringdevicesneedtoattachnasalprobesorchestbandstousers.Wirelesssensingtechnologieshavebeenappliedtomonitorbreathingusingradiowaveswithoutphysicalcontact.Thosewirelesssensingtechnologieshoweverrequirecustomizedradioswhicharenotreadilyavailable.Moreimportantly,duetointerference,suchtechnologiesdonotworkwellwithmultipleusers.Withmultipleusersinpresence,thedetectionaccuracyofexistingsystemsdecreasesdramatically.Inthispaper,weproposetomonitorusersbreathingusingcommercial-off-the-shelf(COTS)RFIDsystems.Inoursystem,passivelightweightRFIDtagsareattachedtousersclothesandbackscatterradiowaves,andcommodityRFIDreadersreportlowleveldata(e.g.,phasevalues).Wetrackperiodicbodymovementduetoinhalingandexhalingbyanalyzingthelowleveldatareportedbycommodityreaders.Toenhancethemeasurementrobustness,wesynthesizedatastreamsfromanarrayofmultipletagstoimprovethemonitoringaccuracy.OurdesignfollowsthestandardEPCprotocolwhicharbitratescollisionsinthepresenceofmultipletags.WeimplementaprototypethebreathmonitoringsystemwithcommodityRFIDsystems.Theexperimentresultsshowthattheprototypesystemcansimultaneouslymonitorbreathingwithhighaccuracyevenwiththepresenceofmultipleusers.
Double-EdgedSword:IncentivizedVerifiableProductPathQueryforRFID-enabledSupplyChain SaiyuQi(XidianUniversity),YuanqingZheng(TheHongKongPolytechnicUniversity),XiaofengChen(XidianUniversity),JianfengMa(XidianUniversity),YongQi(XianJiaotongUniversity)
Queryingthepathinformationofindividualproductsinasupplychainiskeytomanyapplications.RFID(RadioFrequencyIDentification)isamaintechnologytoenableproductpathinformationquerytoday.WithRFIDtechnology,supplychainparticipantscanefficientlytrackproductsintransitandrecordtheirstatesindatabases.Inthispaper,weinvestigatethefollowingquestion:howcanweconductprivacy-preservingproductpathinformationquerywithverifiabilityonanRFIDenableddistributedsupplychain?WeaddressthisquestionwithDoubleEdged(DE)-Sword,anincentivizedverifiablequerysystem.DE-Swordintroducesanoveldouble-edgedreputationincentivemechanismtoencouragesupplychainparticipantstobehave;andcouplesitwithcryptographicprimitivesandcarefulprotocoldesign.WeevaluateDE-Swordthroughsecurityanalysisandperformanceexperiments.ThesecurityanalysisshowsthatDE-Swordguaranteesbothverifiabilityandprivacy.TheexperimentresultsshowthatDE-SwordachieveslowoverheadinRFID-enabledsupplychainapplications.
TowardsAccurateCorruptionEstimationinZigBeeUnderCross-TechnologyInterference GongLongChen(ZhejiangUniversity),WeiDong(ZhejiangUniversity),ZhiweiZhao(UniversityofElectronicScienceandTechnologyofChina),TaoGu(RMITUniversity)
Cross-TechnologyInterferenceaffectstheoperationoflow-powerZigBeenetworks,especiallyundersevereWiFiinterference.AccuratecorruptionestimationisveryimportanttoimprovetheresilienceofZigBeetransmissions.However,therearemanylimitationsinexistingapproachessuchaslowaccuracy,highoverhead,andrequiringhardwaremodification.Inthispaper,weproposeanaccuratecorruptionestimationapproach,AccuEst,whichutilizesper-byteSINR(Signal-to-Interferenceand-NoiseRatio)todetectcorruption.Wecombinetheuseofpilotsymbolswithper-byteSINRtoimprovecorruptiondetectionaccuracy,especiallyinhighlynoisyenvironments(i.e.,noiseandinterferenceareatthesamelevel).Inaddition,wedesignanadaptivepilotinstrumentationschemetostrikeagoodbalancebetweenaccuracyandoverhead.WeimplementAccuEstontheTinyOS2.1.1/TelosBplatformandevaluateitsperformancethroughextensiveexperiments.ResultsshowthatAccuEstimprovescorruptiondetectionaccuracyby78.6%onaveragecomparedwithstate-of-the-artapproach(i.e.,CARE)inhighlynoisyenvironments.Inaddition,AccuEstreducespilotoverheadby53.7%onaveragecomparedtothetraditionalpilot-basedapproach.WeimplementAccuEstinacoding-basedtransmissionprotocol,andresultsshowthatwithAccuEst,thepacketdeliveryratioisimprovedby20.3%onaverage.
UnseenActivityRecognition:AHierarchicalActiveTransferLearningApproach MohammadArifUlAlam(UniversityofMarylandBaltimoreCounty),NirmalyaRoy(UniversityofMarylandBaltimoreCounty)
Humanactivityrecognition(AR)isanessentialelementforuser-centricandcontext-awareapplications.Whilepreviousstudiesshowedpromisingresultsusingvariousmachinelearningalgorithms,mostofthemcanonlyrecognizetheactivitiesthatwerepreviouslyseeninthetrainingdata.Weinvestigatethechallengesofimprovingtherecognitionofunseendailyactivitiesinsmarthomeenvironment,bybetterexploitingthehierarchicaltaxonomyofcomplexdailyactivities.Wefirst(a)designahierarchicalrepresentationofcomplexactivitytaxonomyintermsofhuman-readablesemanticattributes,and(b)developahierarchyofclassifierswhichincorporatesaclustertreebuiltonthedomainknowledgefromtrainingsamples.Thoughthismodelisrichinrecognizingcomplexactivitiesthatare
![Page 63: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/63.jpg)
63
previouslyseenintrainingdata,itisnotwellversedtorecognizeunseencomplexactivitieswithoutnewtrainingsamples.Totacklethischallenge,weextendHierarchicalActiveTransferLearning(HATL)approachthatexploitssemanticattributeclusterstructureofcomplexactivitiessharedbetweenseen(source)andunseen(target)activitydomains.Ourapproachemploystransferandactivelearningtohelplabeltargetdomainunlabeleddatabyspawningthemosteffectivequeries.Weevaluatedourapproachwithtworeal-timesmarthomesystems(IRB#HP-00064387)whichcorroboratesradicalimprovementsinrecognizingunseencomplexactivities.
RFIPad:EnablingCost-efficientandDevice-freeIn-airHandwritingusingPassiveTags HanDing(Xi'anJiaotongUniversity),ChenQian(UniversityofCaliforniaSantaCruz),JinsongHan(Xi'anJiaotongUniversity),GeWang(Xi'anJiaotongUniversity),WeiXi(Xi'anJiaotongUniversity),KunZhao(Xi'anJiaotongUniversity),JizhongZhao(Xi'anJiaotongUniversity)
Animportantfunctionofsmartenvironmentsistheubiquitousaccessofcomputingdevices.Inpublicareassuchashospitals,libraries,andairports,peoplemaywanttointeractwithnearbycomputingsystemstogetinformation,suchasdirectionstoahospitalroom,locationsofbooks,andflightdeparture/arrivalinformation.Touchscreenbaseddisplaysandkiosks,whicharecommonlyusedtoday,mayincurextrahardwarecostorevenpossiblegermandbacteriainfection.Thisworkprovidesanewsolution:userscanmakequeriesandinputsbyperformingin-airhandwritingtoanarrayofpassiveRFIDtags,namedRFIPad.Thisinputmethoddoesnotrequirehumanhandstocarryanydeviceandhenceisconvenientforapplicationsinpublicareas.Besidesthemobileandcontactlessproperty,thissystemisacost-efficientextensiontocurrentRFIDsystems:anexistingreadercanmonitormultipleRFIPadswhileperformingitsregularapplicationssuchasidentificationandtracking.WeimplementaprototypeofRFIPadusingcommercialoff-the-shelfUHFRFIDdevices.ExperimentalresultsshowthatRFIPadachieves>91%accuracyinrecognizingbasictouchscreenoperationsandEnglishletters.
Research8:MobileandWirelessComputingSystemsIRobustIncentiveTreeDesignforMobileCrowdsensing XiangZhang(ArizonaStateUniversity),GuoliangXue(ArizonaStateUniversity),RuozhouYu(ArizonaStateUniversity),DejunYang(ColoradoSchoolofMines),JianTang(SyracuseUniversity)
Withtheproliferationofsmartmobiledevicessuchassmartphones,tablets,andwearable,mobilecrowdsensingbecomesapowerfulsensingandcomputationparadigmwhichhasbeenappliedinmanyfields,suchasspectrumsensing,environmentalmonitoring,healthcare,andsoon.Drivenbypromisingincentives,thepowerofthecrowdgrantscrowdsensinganadvantageinmobilizinguserswhoperformsensingtaskswiththeembeddedsensorsonthesmartdevices.Auctionisoneofthecommonlyadoptedcrowdsensingincentivemechanismstoincentivizeusersforparticipation.However,auctiondoesnotconsidertheincentiveforusersolicitationwhereincrowdsensing,alargenumberofusersisoftenneeded.Todealwiththisissue,weaimtodesignanauction-basedincentivetreetoofferrewardstousersforbothparticipationandsolicitation.Meanwhile,wewanttheincentivemechanismtoberobustagainstdishonestbehaviorsuchasuntruthfulbiddingandsybilattacks,toeliminatethemaliciouspricemanipulation.WedesignanincentivemechanismRIT,whichcombinestheadvantagesofauctionsandincentivetrees.WeprovethatRITistruthfulandsybil-proofwithprobabilityatleastH,foranygivenH_(0,1).WealsoprovethatRITsatisfiesindividualrationality,computationalefficiency,andsolicitationincentive.SimulationresultsofRITfurtherconfirmouranalysis.
WearLock:UnlockingYourPhoneviaAcousticsusingSmartwatch ShanheYi(CollegeofWilliamandMary),ZhengruiQin(NorthwestMissouriStateUniversity),NancyCarter(CollegeofWilliamandMary),QunLi(CollegeofWilliamandMary)
Smartphonelockscreensareimplementedtoreducetheriskofdatalossorcompromisegiventhefactthatincreasingamountofpersondataareaccessibleonsmartphonesnowadays.Unfortunately,manysmartphoneusersabandonlockscreensduetotheinconvenienceofunlockingtheirphonesmanytimesaday.Withthewideadoptionofwearables,token-basedapproacheshavegainedpopularityinsimplifyingunlockingandretainingsecurityatthesametime.Tothisend,weproposetotakeadvantageofthesmartwatchforeasysmartphoneunlocking.Inthispaper,wehavedesignedWearLock,asystemthatusesacoustictonesastokenstoautomatetheunlockingsecurely.Webuildasub-channelselectionandanadaptivemodulationintheacousticmodemtomaximizeunlockingsuccessrateagainstambientnoiseonlywhenthosetwodevicesarenearby.Weleveragethemotionsensoronthesmartwatchtoreducetheunlockfrequency.Weoffloadsmartwatchtaskstothesmartphonetospeedupcomputationandsaveenergy.WehaveimplementedtheWearLockprototypeandconductedextensiveevaluations.Resultsachievedalowaveragebiterrorrate(BER)as8%invariousexperiments.Comparedtotraditionalmanualpersonalidentificationnumbers(PINs)entry,WearLockachievesatleast18%unlockspeedupwithoutanymanualeffort.
ModelingMobileCodeAccelerationintheCloud HuberFlores(UniversityofOulu),XiangSu(UniversityofOulu),VassilisKostakos(UniversityofOulu),JukkaRiekki(UniversityofOulu),EemilLagerspetz(UniversityofHelsinki),SasuTarkoma(HelsinkiUniversityofTechnology),PanHui(HKUST),YongLi(TsinghuaUniversity),JukkaManner(AaltoUniversity)
Tuningthequalityofserviceofamobileapplicationiscriticalinordertoensureuserssatisfaction.Techniqueshavebeenproposedtoaccomplishadaptationofqualityofservicedynamically.However,thereisstillalimitedunderstandingabouthowtoprovideanutilitymodelforcodeexecution.Onekeychallengeistomodelthelevelofqualityinthecodeexecutionthatcanbeprovisionedbythecloud.Sincetheallocationofcloudresourceshasacost,itisimportanttooptimizecloudusage.Weproposeasoftware-definednetworkingapproachthatallowsmodelingandcontrollingcodeaccelerationofamobileapplicationdeployedacrossmultipletypeofdevices.Bysegregatingthecomputationalrequirementsofthemobileapplicationintogroups,wewereabletodefinetheaccelerationneededbyeachgroupofdevices.Asthecomputationalrequirementsofadevicecanchangeacrosstime,amobiledevicecanbere-assignedtoanothergroupbasedondemand.OurSDNapproachimplementsamodelthatallowsthesystemtopredictworkloadbasedonaccelerationgroups.Evaluatingoursysteminarealtestbedshowedthatitispossibletopredictworkloadandallocateoptimalresourcestohandlethatworkloadwith87.5%ofaccuracy.
E-Android:ANewEnergyProfilingToolforSmartphones XingGao(CollegeofWIlliamandMary),DachuanLiu(CollegeofWIlliamandMary),DaipingLiu(UniversityofDelaware),HainingWang(UniversityofDelaware),AngelosStavrou(GeorgeMasonUniversity)
Smartphoneshavebecomeanindispensablepartofourdailylives.Asthelimitedbatterylifetimeremainsamajorfactorrestrictingtheapplicabilityofasmartphone,significantresearcheffortshavebeendevotedtounderstandtheenergyconsumptioninsmartphones.Existingenergymodelingmethodscanaccountenergydraininafine-grainedmannerandprovidewelldesignedhuman-batteryinterfacesforuserstocharacterizeenergyusageofeveryappinsmartphones.
![Page 64: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/64.jpg)
64
However,inthispaper,wedemonstratethattherearestillpitfallsincurrentAndroidenergymodelingapproaches,makingAndroidvulnerabletomaliciousattacks.Inparticular,wepresentasetofnewcollateralenergyattacks,whichcandepletebatterylifebutsidestepthesupervisionofcurrentenergyaccounting.Todefendagainstcollateralenergyattacks,weproposeE-Androidtoaccuratelyprofileenergyconsumptionofasmartphoneinacomprehensivemanner.E-Androidmonitorscollateralenergyrelatedeventsandmaintainsenergyconsumptionmapsforrelevantapps.WeevaluatetheeffectivenessofE-Androidundersixdifferentcollateralenergyattacksandtwonormalscenarios,andcomparetheresultswiththoseofAndroid.WhileAndroidfailstodiscloseallcollateralenergyattacks,E-Androidcanaccuratelyprofileenergyconsumptionandrevealtheexistenceofenergymalware.OurevaluationresultsalsoshowthattheoverheadinducedbyE-Androidisminor.
LocalandLow-CostWhiteSpaceDetection AhmedSaeed(GeorgiaInstituteofTechnology),KhaledHarras(CarnegieMellonUniversity),EllenZegura(GeorgiaInstituteofTechnology),MostafaAmmar(GeorgiaInstituteofTechnology)
WhitespacesareportionsoftheTVspectrumthatareallocatedbutnotusedlocally.Ifaccuratelydetected,whitespacesofferavaluablenewopportunityforhighspeedwirelesscommunications.Weproposeanewmethodforwhitespacedetectionthatallowsanodetoactlocally,basedonacentrallyconstructedmodel,andatlowcost,whiledetectingmorespectrumopportunitiesthanbestknownapproaches.Weleveragetwoideas:first,wedemonstratethatlow-costspectrummonitoringhardwarecanoffergoodenoughdetectioncapabilities.Second,wedevelopamodelthatcombineslocallymeasuredsignalfeaturesandlocationtomoreefficientlydetectwhitespaceavailability.Weincorporatetheseideasintothedesign,implementation,andevaluationofacompletesystemwecallWaldo.WedeployWaldoonlaptopinAtlantametropolitanareaintheUScovering700km2showingthatusingsignalfeaturesinadditiontolocationcanimprovedetectionaccuracybyupto10xforsomechannels.WealsodeployWaldoonanAndroidsmartphone,demonstratingthefeasibilityofreal-timewhitespacedetectionwithefficientuseofsmartphoneresources.
GeneralAnalysisofIncentiveMechanismsforPeer-to-PeerTransmissions:AQuantumGamePerspective WeimanSun(BeijingNormalUniversity),ShenglingWang(BeijingNormalUniversity)
Thepeer-to-peertransmissionisamainstreaminchallengednetworkenvironments.Yet,thefreeriderphenomenoninpeer-topeertransmissionspressesaneedforincentivemechanismstostimulatecontributionsofdatatransmission.Asaresult,itisimperativetoanswerthequestions:whetherandtowhatextentanincentivemechanismcaninvokesuchcontributions?Toanswerthesequestions,weemployann-playercontinuousquantumgamemodeltoanalyzeextrinsicincentivemechanisms(promotingcooperativebehaviorsbyofferingrewards),andusethequantumprisoner’sdilemmamodeltoanalyzeintrinsicincentivemechanisms(encouragingreciprocalcooperationbyexploitinginternalbounds).Tothebestofourknowledge,wearethefirsttoanalyzeincentivemechanismsforpeer-to-peertransmissionsfromaquantumgameperspective.Suchaperspectiveisadoptedbecausetheextendedstrategyspaceinthequantumgamebroadenstherangeforsearchingoptimalstrategiesandtheintroductionofentanglementmakestheproposedanalyticalframeworksmorepracticalduetotheconsiderationofthepeers’relationshipsindecision-making.Ourproposedquantumgame-basedanalyticalframeworksaregenericbecausetheyarecompatiblewithclassicgame-basedschemes.Ouranalyticalresultscanprovidestraightforwardinsightsonevaluatingthepotentialofincentivemechanismsandcanserveasimportantreferencesfordesigningnewincentivemechanisms.
Research9:DistributedBigDataSystemsHigh-PerformanceandResilientKey-ValueStorewithOnlineErasureCodingforBigDataWorkloads DiptiShankar(TheOhioStateUniversity),XiaoyiLu(TheOhioStateUniversity),DhabaleswarPanda(TheOhioStateUniversity)
Distributedkey-valuestore-basedcachingsolutionsarebeingincreasinglyusedtoaccelerateBigDataapplicationsonmodernHPCclusters.Thishasnecessitatedincorporatingfaulttolerancecapabilitiesintohigh-performancekey-valuestoressuchasMemcachedthatareotherwisevolatileinnature.Inmemoryreplicationisbeingusedastheprimarymechanismtoensureresilientdataoperations.However,thisincursincreasednetworkI/Owithhighremotememoryrequirements.Ontheotherhand,erasurecodingisbeingextensivelyexploredforenablingdataresilience,whileachievingbetterstorageefficiency.Inthispaper,wefirstperformanin-depthmodeling-basedanalysisoftheperformancetrade-offsofIn-MemoryReplicationandErasureCodingschemesforkey-valuestores,andexplorethepossibilitiesofemployingOnlineErasureCodingforenablingresilienceinhigh-performancekey-valuestoresforHPCclusters.Wethendesignanon-blockingAPI-basedenginetoperformefficientSet/Getoperationsbyoverlappingtheencoding/decodinginvolvedinenablingErasureCoding-basedresiliencewiththerequest/responsephases,byleveragingRDMAonhighperformanceinterconnects.PerformanceevaluationsshowthattheproposeddesignscanoutperformsynchronousRDMA-basedreplicationbyabout2.8x,andcanimproveYCSBthroughputandaverageread/writelatenciesbyabout1.34x-2.6xoverasynchronousreplicationforlargerkey-valuepairsizes(>16KB).Wealsodemonstrateitsbenefitsbyincorporatingitintoahybridandresilientkey-valuestore-basedburst-buffersystemoverLustreforacceleratingBigDataI/OonHPCclusters.
ModelingandAnalyzingLatencyintheMemcachedsystemWenxueCheng(TsinghuaUniversity),FengyuanRen(TsinghuaUniversity),WanchunJiang(CentralSouthUniversity),TongZhang(TsinghuaUniversity)
Memcachedisawidelyusedin-memorycachingsolutioninlarge-scalesearchingscenarios.ThemostpivotalperformancemetricinMemcachedislatency,whichisaffectedbyvariousfactorsincludingtheworkloadpattern,theservicerate,theunbalancedloaddistributionandthecachemissratio.Toquantitatetheimpactofeachfactoronlatency,weestablishatheoreticalmodelfortheMemcachedsystem.Specially,weformulatetheunbalancedloaddistributionamongMemcachedserversbyasetofprobabilities,capturetheburstandconcurrentkeyarrivalsatMemcachedserversinformofbatchingblocks,andaddacachemissprocessingstage.Basedonthismodel,algebraicderivationsareconductedtoestimatelatencyinMemcached.Thelatencyestimationisvalidatedbyintensiveexperiments.Moreover,weobtainaquantitativeunderstandingofhowmuchimprovementoflatencyperformancecanbeachievedbyoptimizingeachfactorandprovideseveralusefulrecommendationstooptimallatencyinMemcached.
SpeculativeSlotReservation:EnforcingServiceIsolationforDependentData-ParallelComputations ChenChen(HKUST),WeiWang(HKUST),BoLi(HKUST)
Priorityschedulingisafundamentaltooltoprovideserviceisolationfordifferentjobsinsharedclusters.Ideally,theperformanceofahigh-priorityjobshouldnotbedraggeddownbyanotherwithalowerpriority.However,weshowinthispaperthatsimplyassigningahighpriorityprovidesnoisolationforjobswithdependentcomputations.Ajob,evenreceivingthehighestpriority,maygiveupcomputeslotstoanotherbeforeproceedingtothedownstreamcomputation,whichisbecause
![Page 65: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/65.jpg)
65
ofbarrier,i.e.,thatthedownstreamcomputationcannotstartuntilalltheupstreamtaskshavecompleted.Suchaninterruptionofexecutioninevitablyresultsinasignificantdelay.Inthispaper,weproposespeculativeslotreservationthatjudiciouslyreservesslotsfordownstreamcomputations,soastoretainserviceisolationforhigh-priorityjobs.Tomitigatetheutilizationlossduetoslotreservation,weanalyzethetrade-offbetweenutilizationandisolation,andexposeatunableknobtonavigatethetrade-off.Wealsoproposeacomplementarystragglermitigationstrategythatusesthereservedslotstorunextracopiesofslowtasks.WehaveimplementedspeculativeslotreservationinSpark.Evaluationsbasedonbothclusterdeploymentandtrace-drivensimulationsshowthatourapproachenforcesstrictserviceisolationforhigh-priorityjobs,withoutslowingdowntheotherjobswithalowerpriority.
OptimizingShuffleinWide-AreaDataAnalytics ShuhaoLiu(UniversityofToronto),HaoWang(UniversityofToronto),BaochunLi(UniversityofToronto)
Asincreasinglylargevolumesofrawdataaregeneratedatgeographicallydistributeddatacenters,theyneedtobeefficientlyprocessedbydataanalyticjobsspanningmultipledatacentersacrosswide-areanetworks.Designedforasingledatacenter,existingdataprocessingframeworks,suchasApacheSpark,arenotabletodeliversatisfactoryperformancewhenthesewide-areaanalyticjobsareexecuted.Aswide-areanetworksinterconnectingdatacentersmaynotbecongestionfree,thereisacompellingneedforanewsystemframeworkthatisoptimizedforwide-areadataanalytics.Inthispaper,wedesignandimplementanewproactivedataaggregationframeworkbasedonApacheSpark,withafocusonoptimizingthenetworktrafficincurredinshufflestagesofdataanalyticjobs.Theobjectiveofthisframeworkistostrategicallyandproactivelyaggregatetheoutputdataofmappertaskstoasubsetofworkerdatacenters,asareplacementtoSparksoriginalpassivefetchmechanismacrossdatacenters.Itimprovestheperformanceofwide-areaanalyticjobsbyavoidingrepetitivedatatransfers,whichimprovestheutilizationofinter-datacenterlinks.OurextensiveexperimentalresultsusingstandardbenchmarksacrosssixAmazonEC2regionshaveshownthatourproposedframeworkisabletoreducejobcompletiontimesbyupto73%,ascomparedtotheexistingbaselineimplementationinSpark.
JobSchedulingwithoutPriorInformationinBigDataProcessingSystems ZhimingHu(UniversityofToronto),BaochunLi(UniversityofToronto),ZhengQin(InstituteofHighPerformanceComputing),RickSiowMongGoh(InstituteofHighPerformanceComputing)
Jobschedulingplaysanimportantroleforimprovingtheoverallsystemperformanceinbigdataprocessingframeworks.Simplejobschedulingpolicies,suchasFairandFIFOschedulingdonotconsiderjobsizes,andmaydegradetheperformancewhenjobsofvaryingsizesarrive.Moreelaboratejobschedulingpoliciesmaketheconvenientassumptionthatjobsarerecurring,andcompleteinformationabouttheirsizesisavailablefromtheirpriorruns.Inthispaper,wedesignandimplementanefficientandpracticaljobschedulerforbigdataprocessingsystemstoachievebetterperformanceevenwithoutpriorinformationaboutjobsizes.Thesuperiorperformanceofourjobscheduleroriginatesfromthedesignofamultiplelevelpriorityqueue,wherejobsaredemotedtolowerpriorityqueuesiftheamountofserviceconsumedsofarreachesacertainthreshold.Inthiscase,jobsinneedofasmallamountofservicecanfinishinthetopmostseverallevelsofqueues,whilejobsthatneedalargeamountofservicetocompletearemovedtolowerpriorityqueuestoavoidhead-of-lineblocking.Ournewjobschedulercaneffectivelymimicashortestjobfirstschedulingpolicywithoutknowingthejobsizesinadvance.Todemonstrateitsperformance,wehaveimplementedournewjobschedulerinYARN,apopularresourcemanagerusedbyHadoop/Spark,andvalidateditsperformancewithbothexperimentsonrealdatasetsandlarge-scaletrace-drivensimulations.Ourexperimentalandsimulationresultshavestronglyconfirmedtheeffectivenessofourdesign:ournewjobschedulercanreducetheaveragejobresponsetimeoftheFairschedulerbyupto45%.
DistributedLoadBalancinginKey-ValueNetworkedCachesSikderHuq(TheUniversityofIowa),ZubairShafiq(TheUniversityofIowa),SukumarGhosh(TheUniversityofIowa),AmirKhakpour(VerizonDigitalMediaServices),HarkeeratBedi(VerizonDigitalMediaServices)
Modernwebservicesrelyonanetworkofdistributedcacheserverstoefficientlydelivercontenttousers.Loadimbalanceamongcacheserverscansubstantiallydegradecontentdeliveryperformance.Duetotheskewedanddynamicnatureofreal-worldworkloads,cacheserversthatserveviralcontentexperiencehigherloadascomparedtoothercacheservers.WeproposeanoveldistributedloadbalancingprotocolcalledMeezantoaddresstheloadimbalanceamongcacheservers.Meezanreplicatespopularobjectstomitigateskewnessandadjustshashspaceboundariesinresponsetoloaddynamicsinanovelway.OurtheoreticalanalysisshowsthatMeezanachievesnearperfectloadbalancingforawiderangeofoperatingparameters.OurtracedrivensimulationsshowsthatMeezanreducesloadimbalancebyupto52%ascomparedtopriorsolutions.
Research10:DistributedAlgorithmsandTheoryICognitiveContext-awareDistributedStorageOptimizationinMobileCloudComputing:AStableMatchingbasedApproach DongHan(OaklandUniversity),YeYan(OaklandUniversity),TaoShu(AuburnUniversity),LiuqingYang(ColoradoStateUniversity),ShuguangCui(UniversityofCalifornia,Davis)
Mobilecloudstorage(MCS)isbeingextensivelyusednowadaystoprovidedataaccessservicestovariousmobileplatformssuchassmartphonesandtablets.Forcross-platformmobileapps,MCSisafoundationforsharingandaccessinguserdataaswellassupportingseamlessuserexperienceinamobilecloudcomputingenvironment.However,themobileusageofsmartphonesortabletsisquitedifferentfromlegacydesktopcomputers,inthesensethateachuserhashis/herownmobileusagepattern.Therefore,itischallengingtodesignanefficientMCSthatisoptimizedforindividualusers.Inthispaper,weinvestigateadistributedMCSsystemwhoseperformanceisoptimizedbyexploitingthefine-grainedcontextinformationofeverymobileuser.Inthisdistributedsystem,lightweightstorageserversaredeployedpervasively,suchthatdatacanbestoredclosertoitsuser.WesystematicallyoptimizethedataaccessefficiencyofsuchadistributedMCSbyexploitingthreetypesofusercontextinformation:mobilitypattern,networkcondition,anddataaccesspattern.Weproposetwooptimizationformulations:acentralizedonebasedonmixed-integerlinearprogramming(MILP),andadistributedonebasedonstablematching.Wethendevelopsolutionstobothformulations.Comprehensivesimulationsareperformedtoevaluatetheeffectivenessoftheproposedsolutionsbycomparingthemagainsttheircounterpartsundervariousnetworkandcontextconditions.
FairCachingAlgorithmsforPeerDataSharinginPervasiveEdgeComputingEnvironments YaodongHuang(StonyBrookUniversity),XintongSong(PekingUniversity),FanYe(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity),XiaomingLi(PekingUniversity)
Edgedevices(e.g.,smartphones,tablets,connectedvehicles,IoTnodes)withsensing,storageandcommunicationresourcesareincreasinglypenetratingourenvironments.Manynovelapplicationscanbecreatedwhennearbypeeredgedevicessharedata.Cachingcangreatlyimprovethedataavailability,retrieval
![Page 66: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/66.jpg)
66
robustnessandlatency.Inthispaper,westudytheuniqueissueofcachingfairnessinedgeenvironment.Duetodistinctownershipofpeerdevices,cachingloadbalanceiscritical.Weconsiderfairnessmetricsandformulateanintegerlinearprogrammingproblem,whichisshownassummationofmultipleConnectedFacilityLocation(ConFL)problems.WeproposeanapproximationalgorithmleveraginganexistingConFLapproximationalgorithm,andprovethatitpreservesa6.55approximationratio.Wefurtherdevelopadistributedalgorithmwheredevicesexchangedatareachabilityandidentifypopulatecandidatesascachingnodes.Extensiveevaluationshowsthatcomparedwithexistingwirelessnetworkcachingalgorithms,ouralgorithmsimprovesthe75-percentilefairnessfrom22.8%to71.4%,whileachievingcontentionthuslatencysimilarasthebestexistingworks.
Latency-DrivenCooperativeTaskComputinginMulti-UserFog-RadioAccessNetworks Ai-ChunPang(NationalTaiwanUniversity),Wei-HoChung(AcademiaSinica),Te-ChuanChiu(NationalTaiwanUniversity),JunshanZhang(ArizonaStateUniversity)
Fogcomputingisemergingasonepromisingsolutiontomeettheincreasingdemandforultra-lowlatencyservicesinwirelessnetworks.Takingaforward-lookingperspective,weproposeaFog-RadioAccessNetwork(F-RAN)model,whichutilizestheexistinginfrastructure,e.g.,smallcellsandmacrobasestations,toachievetheultra-lowlatencybyjointcomputingacrossmultipleF-RANnodesandnear-rangecommunicationsattheedge.Wetreatthelowlatencydesignasanoptimizationproblem,whichcharacterizesthetradeoffbetweencommunicationandcomputingacrossmultipleF-RANnodes.SincethisproblemisNP-hard,weproposealatency-drivencooperativetaskcomputingalgorithmwithone-for-allconceptforsimultaneousselectionoftheF-RANnodestoservewithproperheterogeneousresourceallocationformulti-userservices.Consideringthelimitedheterogeneousresourcessharedamongallusers,weadvocatetheone-for-allstrategyforeveryusertakingotherssituationintoconsiderationandseekforawinwinsolution.Thenumericalresultsshowthatthelow-latencyservicescanbeachievedbyF-RANvialatency-drivencooperativetaskcomputing.
ApproximationandOnlineAlgorithmsforNFV-EnabledMulticastinginSDNs ZichuanXu(UniversityCollegeLondon),WeifaLiang(TheAustralianNationalUniversity),MeitianHuang(TheAustralianNationalUniversity),MikeJia(TheAustralianNationalUniversity),SongGuo(TheHongKongPolytechnicUniversity),AlexGalis(UniversityCollegeLondon)
Multicastingisafundamentalfunctionalityofnetworksformanyapplicationsincludingonlineconferencing,eventmonitoring,videostreaming,andsystemmonitoringindatacenters.Toensuremulticastingreliable,secureandscalable,aservicechainconsistingofnetworkfunctions(e.g.,firewalls,IntrusionDetectionSystems(IDSs),andtranscoders)usuallyisassociatedwitheachmulticastrequest.SuchamulticastrequestisreferredtoasanNFV-enabledmulticastrequest.InthispaperwestudyNFV-enabledmulticastinginaSoftware-DefinedNetwork(SDN)withtheaimstominimizetheimplementationcostofeachNFV-enabledmulticastrequestormaximizethenetworkthroughputforasequenceofNFV-enabledrequests,subjecttonetworkresourcecapacityconstraints.WefirstformulatenovelNFV-enabledmulticastingandonlineNFV-enabledmulticastingproblems.Wethendevisetheveryfirstapproximationalgorithmwithanapproximationratioof2KfortheNFV-enabledmulticastingproblemifthenumberofserversforimplementingthenetworkfunctionsofeachrequestisnomorethanaconstantK(1).WealsostudydynamicadmissionsofNFV-enabledmulticastrequestswithouttheknowledgeoffuturerequestarrivalswiththeobjectivetomaximizethenetworkthroughput,forwhichweproposeanonlinealgorithmwithacompetitiveratioofO(logn)whenK=1,wherenisthenumberofnodesinthenetwork.Wefinallyevaluatetheperformanceoftheproposedalgorithmsthroughexperimentalsimulations.Experimentalresultsdemonstratethattheproposedalgorithmsoutperformotherexistingheuristics.
DistributedAuctionsforTaskAssignmentandSchedulinginMobileCrowdsensingSystems ZhuojunDuan(GeorgiaStateUniversity),WeiLi(GeorgiaStateUniversity),ZhipengCai(GeorgiaStateUniversity)
WiththeemergenceofMobileCrowdsensingSystems(MCSs),manyauctionschemeshavebeenproposedtoincentivizemobileuserstoparticipateinsensingactivities.However,inmostoftheexistingwork,theheterogeneityofMCSshasnotbeenfullyexploited.Totacklethisissue,inthispaper,westudythejointproblemofsensingtaskassignmentandschedulingwhileconsideringpartialfulfillment,attributediversity,andpricediversity.Wefirstelaboratelymodeltheproblemasareverseauctionanddesignadistributedauctionframework.Then,basedonthisframework,weproposetwodistributedauctionschemes,costpreferredauctionscheme(CPAS)andtimeschedule-preferredauctionscheme(TPAS),whichdifferonthemethodsoftaskscheduling,winnerdetermination,andpaymentcomputation.WefurtherrigorouslyprovethatbothCPASandTPAScanachievecomputational-efficiency,individual-rationality,budgetbalance,andtruthfulness.Finally,thesimulationresultsvalidatetheeffectivenessofbothCPASandTPASintermsofsensingtasksallocationefficiency,mobileusersworkingtimeutilizationandutility,andtruthfulness.
EffectiveMobileDataTradinginSecondaryAd-hocMarketwithHeterogeneousandDynamicEnvironment HengkySusanto(HuaweiFutureNetworkTheoryLab),HonggangZhang(UniversityofMassachusettsBoston),ShingYipHo(ShareMedia),BenyuanLiu(UniversityofMassachusettsLowell)
Advancesinsmartphonetechnologiesenablemobiledatasubscriberstoreselltheirdataallowancetootherusers,creatingasecondarydatamarket.Thetradingenvironmentofthissecondarydatamarketisdynamicandad-hoc:buyersandsellersjoinandleavethemarketatalltimes,changingthetradinglandscapeconstantly.Theamountofdatademandedandofferedatanypointintimealsovary.Theseconditionsmakedeterminingafairtransactionprice,andmatchingbuyerstosellersdifficultinpractice.Priorschemesutilizeglobaldescriptionofthenetworkandmarketforcestoachievegoodperformance,buttheimplementationrequiresahighoverheadcost.Inthispaper,wepresentDataMart,adatapricingandusermatchingplatformfortradinginthisdynamic,ad-hocandheterogeneousmarketthatworksindistributedmannerwithoutneedingglobalinformation.Usinginsightsfromrealworldtraces,wedemonstrateviasimulationthatourpricingschemeisconvergingandconsistentwiththelawofdemandandsupply.Further,ourusermatchingschemeachievescomparableperformancetotheoptimalsolution.WeimplementaprototypeonAndroidplatform,andtheexperimentresultsconfirmtheeffectivenessofDataMart.
Research11:SecurityandPrivacyinDistributedSystemsIIKalis-ASystemforKnowledge-drivenAdaptableIntrusionDetectionfortheInternetofThings DanieleMidi(PurdueUniversity),AntoninoRullo(UniversityofCalabria),AnandMudgerikar(PurdueUniversity),ElisaBertino(PurdueUniversity)
Inthispaper,weintroduceKalis,aself-adapting,knowledge-drivenexpertIntrusionDetectionSystemabletodetectattacksinrealtimeacrossawiderangeofIoTsystems.KalisdoesnotrequirechangestoexistingIoTsoftware,canmonitorawidevarietyofprotocols,hasnoperformanceimpactonapplicationsonIoTdevices,andenablescollaborativesecurityscenarios.KalisisthefirstcomprehensiveapproachtointrusiondetectionforIoTthatdoesnottargetindividualprotocolsor
![Page 67: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/67.jpg)
67
applications,andadaptsthedetectionstrategytothespecificnetworkfeatures.ExtensiveevaluationshowsthatKalisiseffectiveandefficientindetectingattackstoIoTsystems.
FuzzyExtractorsforBiometricIdentification NanLi(CSIRO),FuchunGuo(UniversityofWollongong),YiMu(UniversityofWollongong),WillySusilo(UniversityofWollongong),SuryaNepal(CSIRO)
Fuzzyextractorprovideskeygenerationfrombiometricsandothernoisydata.Thegeneratedkeyisseamlesslyusableforanycryptographicapplicationsbecauseitsinformationentropyissufficientforsecurity.Biometricauthenticationoffersnaturalandpasswordlessuserauthenticationinvarioussystemswherefuzzyextractorscanbeusedforbiometricinformationsecurity.Typically,abiometricsystemoperatesintwomodes:verificationandidentification.However,existingfuzzyextractorsdoesnotsupportefficientuseridentification.Inthispaper,weproposeasuccinctfuzzyextractorschemewhichenablesefficientbiometricidentificationaswellasverificationanditsatisfiessecurityrequirements.Weshowthattheproposedschemecanbeeasilyusedinbothverificationandidentificationmodes.Tothebestofourknowledge,weproposethefirstfuzzyextractorbasedbiometricidentificationprotocol.Theproposedprotocolisabletoidentifyauserwithconstantcomputationalcostratherthanlinear-timecomputationbyusingotherfuzzyextractorschemes.Biometricinformationsecurityisanimportantconcernofusingbiometricsystems,sothatwealsoprovidesecurityanalysisofproposedschemestoshowtheirsecurityboundaries.Theimplementationshowsthattheperformanceofproposedidentificationprotocolisconstantandclosetothatofverificationprotocols.
SmartphonePrivacyLeakageofSocialRelationshipsandDemographicsfromSurroundingAccessPoints ChenWang(StevensInstituteofTechnology),ChuyuWang(StevensInstituteofTechnology),YingyingChen(StevensInstituteofTechnology),LeiXie(NanjingUniversity),SangluLu(NanjingUniversity)
WhilethemobileusersenjoytheanytimeanywhereInternetaccessbyconnectingtheirmobiledevicesthroughWi-Fiservices,theincreasingdeploymentofaccesspoints(APs)haveraisedanumberofprivacyconcerns.ThispaperexploresthepotentialofsmartphoneprivacyleakagecausedbysurroundingAPs.Inparticular,westudytowhatextenttheuserspersonalinformationsuchassocialrelationshipsanddemographicscouldberevealedleveragingsimplesignalinformationfromAPswithoutexaminingtheWi-Fitraffic.OurapproachutilizesusersactivitiesatdailyvisitedplacesderivedfromthesurroundingAPstoinferuserssocialinteractionsandindividualbehaviors.Furthermore,wedeveloptwonewmechanisms:theCloseness-basedSocialRelationshipsInferencealgorithmcaptureshowcloselypeopleinteractwitheachotherbyevaluatingtheirphysicalclosenessandderivesfine-grainedsocialrelationships,whereastheBehavior-basedDemographicsInferencemethoddifferentiatesvariousindividualbehaviorsviatheextractedactivityfeatures(e.g.,activenessandtimeslots)ateachdailyplacetorevealusersdemographics.Extensiveexperimentsconductedwith21participantsrealdailylifeincluding257differentplacesinthreecitiesovera6-monthperioddemonstratethatthesimplesignalinformationfromsurroundingAPshaveahighpotentialtorevealpeoplessocialrelationshipsandinferdemographicswithanover90%accuracywhenusingourapproach.
EV-Matching:BridgingLargeVisualDataandElectronicDataforEfficientSurveillanceGangLi(TheOhioStateUniversity),FanYang(TheOhioStateUniversity),GuoxingChen(TheOhioStateUniversity),QiangZhai(TheOhioStateUniversity),XinfengLi(TheOhioStateUniversity),JinTeng(TheOhioStateUniversity),JundaZhu(UniversityofMacau),DongXuan(TheOhioStateUniversity),BiaoChen(UniversityofMacau),WeiZhao(UniversityofMacau)
Visual(V)surveillancesystemsareextensivelydeployedandbecomingthelargestsourceofbigdata.Ontheotherhand,electronic(E)dataalsoplaysanimportantroleinsurveillanceanditsamountincreasesexplosivelywiththeubiquityofmobiledevices.Oneofthemajorproblemsinsurveillanceistodeterminehumanobjectsidentitiesamongdifferentsurveillancescenes.TraditionalwayofprocessingbigVandEdatasetsseparatelydoesnotservethepurposewellbecauseVdataandEdataareimperfectaloneforinformationgatheringandretrieval.Matchinghumanobjectsinthetwodatasetscanmergethegoodofthetwoforefficientlarge-scalesurveillance.Yetsuchmatchingacrosstwoheterogeneousbigdatasetsischallenging.Inthispaper,weproposeanefficientsetofparallelalgorithms,calledEV-Matching,tobridgebigEandVdata.WematchEandVdatabasedontheirspatiotemporalcorrelation.TheEV-MatchingalgorithmsareimplementedonApacheSparktofurtheracceleratethewholeprocedure.Weconductextensiveexperimentsonalargesyntheticdatasetunderdifferentsettings.Resultsdemonstratesthefeasibilityandefficiencyofourproposedalgorithms.
AdaptiveReconnaissanceAttackswithNear-OptimalParallelBatching XiangLi(UniversityofFlorida),JohnathanSmith(UniversityofFlorida),MyThai(UniversityofFlorida)
Inassessingprivacyononlinesocialnetworks,itisimportanttoinvestigatetheirvulnerabilitytoreconnaissancestrategies,inwhichattackersluretargetsintobeingtheirfriendsbyexploitingthesocialgraphinordertoextractvictimssensitiveinformation.Asthenetworktopologyisonlypartiallyrevealedaftereachsuccessfulfriendrequest,attackersneedtoemployanadaptivestrategy.Existingworkonlyconsideredasimplestrategyinwhichattackerssequentiallyacquireonefriendatatime,whichcausestremendousdelayinwaitingforresponsesbeforesendingthenextrequest,andwhichlacktheabilitytoretryfailedrequestsafterthenetworkhaschanged.Incontrast,weinvestigateanadaptiveandparallelstrategy,ofwhichattackerscansimultaneouslysendmultiplefriendrequestsinbatchandrecoverfromfailedrequestsbyretryingaftertopologychanges,therebysignificantlyreducingthetimeofreachingthetargetsandimprovingrobustness.Wecastthisapproachasanoptimizationproblem,Max-Crawling,andshowitinapproximablewithin(1_1/e+).WefirstdesignourcorealgorithmPM-AReSTwhichhasanapproximationratioof(1_e(1/e_1))usingadaptivemonotonicsubmodularproperties.Wenextprovideanear-optimalsolution((1_1/e))viaatwostagestochasticprogrammingapproach.Wefurtherestablishthegapboundof(1_e_(1_1/e)2)betweenbatchstrategiesversustheoptimalsequentialone.Weexperimentallyvalidateourtheoreticalresults,findingthatouralgorithmperformsnearoptimallyinpracticeandthatthisisrobustunderavarietyofproblemsettings.
AchievingStrongPrivacyinOnlineSurvey YouZhou(UniversityofFlorida),YianZhou(UniversityofFlorida),ShigangChen(UniversityofFlorida),SamuelS.Wu(UniversityofFlorida)
ThankstotheproliferationofInternetaccessandmoderndigitalandmobiledevices,onlinesurveyhasbeenflourishingintodatacollectionofmarketing,social,financialandmedicalstudies.However,traditionaldatacollectionmethodsinonlinesurveysufferfromseriousprivacyissues.Existingprivacyprotectiontechniquesarenotadequateforonlinesurveyforlackofstrongprivacy.Inthispaper,weproposeapracticalstrongprivacyonlinesurveyschemeSPSbasedonanoveldatacollectiontechniquecalleddualmatrixmasking(DM2),whichguaranteesthecorrectnessofthetallyingresultswithlowcomputationoverhead,andachievesuniversalverifiability,robustnessandstrongprivacy.WealsoproposeamorerobustschemeRSPS,whichincorporatesmultipledistributedsurveymanagers.The
![Page 68: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/68.jpg)
68
RSPSschemepreservesthenicepropertiesofSPS,andfurtherachievesrobuststrongprivacyagainstjointcollusionattack.Throughextensiveanalyses,wedemonstrateourproposedschemescanbeefficientlyappliedtoonlinesurveywithaccuracyandstrongprivacy.
Research12:CloudComputingandDistributedDataAnalyticsServiceOverlayForestEmbeddingforSoftware-DefinedCloudNetworks Jian-JhihKuo(AcademiaSinica),Shan-HsiangShen(NationalTaiwanUniversityofScienceandTechnology),Ming-HongYang(UniversityofMinnesota),De-NianYang(AcademiaSinica),Ming-JerTsai(NationalTsingHuaUniversity),Wen-TsuenChen(AcademiaSinica)
NetworkFunctionVirtualization(NFV)onSoftwareDefinedNetworks(SDN)caneffectivelyoptimizetheallocationofVirtualNetworkFunctions(VNFs)andtheroutingofnetworkflowssimultaneously.Nevertheless,mostpreviousstudiesonNFVfocusonunicastservicechainsandtherebyarenotscalabletosupportalargenumberofdestinationsinmulticast.Ontheotherhand,theallocationofVNFshasnotbeensupportedinthecurrentSDNmulticastroutingalgorithms.Inthispaper,therefore,wemakethefirstattempttotackleanewchallengingproblemforfindingaserviceforestwithmultipleservicetrees,whereeachtreecontainsmultipleVNFsrequiredbyeachdestination.Specifically,weformulateanewoptimization,namedServiceOverlayForest(SOF),tominimizethetotalcostofallallocatedVNFsandallmulticasttreesintheforest.Wedesignanew3_ST-approximationalgorithmtosolvetheproblem,where_STdenotesthebestapproximationratiooftheSteinerTreeproblem,andthedistributedimplementationofthealgorithmisalsopresented.Simulationresultsonrealnetworksfordatacentersmanifestthattheproposedalgorithmoutperformstheexistingonesbyover25%.Moreover,theimplementationofanexperimentalSDNwithHPOpenFlowswitchesindicatesthatSOFcansignificantlyimprovetheQoEoftheYoutubeservice.
JointOptimizationofChainPlacementandRequestSchedulingforNetworkFunctionVirtualization QixiaZhang(HuazhongUniversityofScience&Technology),YikaiXiao(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),JohnChiShingLui(ChineseUniversityofHongKong),JianGuo(HuazhongUniversityofScience&Technology),TaoWang(HuazhongUniversityofScience&Technology)
ComparedwithexecutingNetworkFunctions(NFs)ondedicatedhardwares,therecenttrendofNetworkFunctionVirtualization(NFV)holdsthepromiseforoperatorstoflexiblydeploysoftware-basedNFsoncommodityservers.However,virtualNFs(VNFs)arenormallychainedtogethertoprovideaspecificnetworkservice.Thus,anefficientschemeisneededtoplacetheVNFchainsacrossthenetworkandeffectivelyschedulerequeststoserviceinstances,whichcanmaximizetheaverageresourceutilizationofeachnodeinserviceandsimultaneouslyminimizetheaverageresponselatencyofeachrequest.Tothisend,weformulatefirstVNFchainsplacementproblemasavariantofbin-packingproblem,whichisNP-hard,andwemodelrequestschedulingproblembasedonthekeyconceptsfromopenJacksonnetwork.TojointlyoptimizetheperformanceofNFV,weproposeapriority-drivenweightedalgorithmtoimproveresourceutilizationandaheuristicalgorithmtoreduceresponselatency.Throughextensivetrace-drivensimulations,weshowthatourmethodscanindeedenhanceperformanceindiversescenarios.Inparticular,wecanimprovetheaverageresourceutilizationby24.9%andcanreducetheaveragetotallatencyby19.9%ascomparedwithotherstate-of-the-artmethods.
BIGCacheAbstractionforCacheNetworks EmanRamadan(UniversityofMinnesota),ArvindNarayanan(UniversityofMinnesota),Zhi-LiZhang(UniversityofMinnesota),RunhuiLi(HuaweiFutureNetworkTheoryLab),GongZhang(HuaweiFutureNetworkTheoryLab)
Inthispaper,weadvocatethenotionofBIGcacheasaninnovativeabstractionforeffectivelyutilizingthedistributedstorageandprocessingcapacitiesofallserversinacachenetwork.TheBIGcacheabstractionisproposedtopartlyaddresstheproblemof(cascade)thrashinginahierarchicalnetworkofcacheservers,whereithasbeenknownthatcacheresourcesatintermediateserversarepoorlyutilized,especiallyunderclassicalcachereplacementpoliciessuchasLRU.WelayouttheadvantagesofBIGcacheabstractionandmakeastrongcasebothfromatheoreticalstandpointaswellasthroughsimulationanalysis.WealsodevelopthedCLIMBcachealgorithmtominimizetheoverheadsofmovingobjectsacrossdistributedcacheboundariesandpresentasimpleyeteffectiveheuristicforaddressingthecacheallotmentprobleminthedesignofBIGcacheabstraction.
DistributedQRdecompositionframeworkfortrainingSupportVectorMachines JyotikrishnaDass(TexasA&MUniversity),V.N.S.PrithviSakuru(TexasA&MUniversity),VivekSarin(TexasA&MUniversity),RabiN.Mahapatra(TexasA&MUniversity)
SupportVectorMachines(SVM)belongtoclassofsupervisedmachinelearningalgorithmswithapplicationsinclassificationandregressionanalysis.SVMtrainingismodeledasaconvexoptimizationproblemthatiscomputationallytediousandhaslargememoryrequirements.Specifically,itisaquadraticprogrammingproblemwhichscalesrapidlywiththetrainingsetsizeratherthanthedimensionalityofthefeaturespace.Inthiswork,wefirstpresentanovelQRdecompositionframework(QRSVM)toefficientlymodelandsolvealargescaleSVMproblembycapitalizingonlow-rankrepresentationsofthefullkernelmatrixratherthansolvingtheproblemasasequenceofsmallersub-problems.Thelow-rankstructureofthekernelmatrixisleveragedtotransformthedensematrixintoonewithasparseandseparablestructure.ThemodifiedSVMproblemrequiressignificantlylessermemoryandcomputation.Ourapproachscaleslinearlywiththetrainingsetsizewhichmakesitapplicabletolargedatasets.Thismotivatestowardsouranothercontribution;exploringadistributedQRSVMframeworktosolvelarge-scaleSVMclassificationproblemsinparallelacrossaclusterofcomputingnodes.Wealsoderiveanoptimalstepsizeforfastconvergenceofthedualascentmethodwhichisusedtosolvethequadraticprogrammingproblem.
DistributivelyComputingRandomWalkBetweennessCentralityinLinearTime Qiang-ShengHua(HuazhongUniversityofScienceandTechnology),MingAi(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),DongxiaoYu(HuazhongUniversityofScienceandTechnology),XuanhuaShi(HuazhongUniversityofScienceandTechnology)
Betweennesscentralityofanoderepresentsitsinfluenceoverthespreadofinformationinthenetwork.Itisnormallydefinedastheratioofthenumberofshortestpathspassingthroughthenodeamongallshortestpaths.However,thespreadofinformationmaynotjustpassthroughtheshortestpathswhichiscapturedbyanewmeasureofbetweennesscentralitybasedonrandomwalks[1].Therandomwalkbetweennesscentralityofanodemeanshowoftenitistraversedbyarandomwalkbetweenallpairsofothernodes.Inthispaper,weproposeanO(nlogn)timedistributedrandomizedapproximationalgorithmforcalculatingeachnodesrandomwalkbetweennesscentralitywithanapproximationratio(1_)wherenisthenumberofnodesandisanarbitrarilysmallconstantbetween0and1.OurdistributedalgorithmisdesignedunderthewidelyusedCONGESTmodel,whereeachedgecanonlytransferO(logn)bitsineachround.Toourbest
![Page 69: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/69.jpg)
69
knowledge,thisisthefirstdistributedalgorithmforcomputingtherandomwalkbetweennesscentrality.Moreover,wegiveanon-triviallowerboundfordistributivelycomputingtheexactrandomwalkbetweennesscentralityundertheCONGESTmodel,whichis(nlogn+D)whereDisthenetworkdiameter.Thismeansexactlycomputingrandomwalkbetweennesscannotbedoneinsublineartime.
DeGPar:LargeScaleTopicDetectionusingNode-CutPartitioningonDenseWeightedGraphs KambizGhoorchian(RoyalInstituteofTechnology(KTH)),SarunasGirdzijauskas(RoyalInstituteofTechnology(KTH)),FatemehRahimian(SwedishInstituteofComputerScience(SICS))
TopicDetection(TD)referstoautomatictechniquesforlocatingtopicallyrelatedmaterialinwebdocuments[1].Nowadays,massiveamountsofdocumentsaregeneratedbyusersofOnlineSocialNetworks(OSNs),informofveryshorttext,tweetsandsnippetsofnews.Whiletopicdetection,initstraditionalform,isappliedtoafewdocumentscontainingalotofinformation,theproblemhasnowchangedtodealingwithmassivenumberofdocumentswithverylittleinformation.Thetraditionalsolutions,thus,fallshorteitherinscalability(duetohugenumberofinputitems)orsparsity(duetoinsufficientinformationperinputitem).InthispaperweaddressthescalabilityproblembyintroducinganefficientandscalablegraphbasedalgorithmforTDonshorttexts,leveragingdimensionalityreductionandclusteringtechniques.Wefirst,compresstheinputsetofdocumentsintoadensegraph,suchthatfrequentcooccurrencepatternsinthedocumentscreatemultipledensetopologicalareasinthegraph.Then,wepartitionthegraphintomultipledensesub-graphs,eachrepresentingatopic.Wecomparetheaccuracyandscalabilityofoursolutionwithtwostate-of-the-artsolutions(includingthestandardLDA,andBiTerm).Theresultsontwowidelyusedbenchmarkdatasetsshowthatouralgorithmnotonlymaintainsasimilarorbetteraccuracy,butalsoperformsbyanorderofmagnitudefasterthanthestate-of-the-artapproaches.
Research13:DistributedAlgorithmsandTheoryIINetworkedStochasticMulti-ArmedBanditswithCombinatorialStrategies ShaojieTang(UniversityofTexasatDallas),YaqinZhou(SUTD),KaiHan(UniversityofScienceandTechnologyofChina),ZhaoZhang(ZhejiangNormalUniversity),JingYuan(UniversityofTexasatDallas),WeiliWu(UniversityofTexasatDallas)
Inthispaper,weinvestigatealargelyextendedversionofclassicalMABproblem,callednetworkedcombinatorialbanditproblems.Inparticular,weconsiderthesettingofadecisionmakeroveranetworkedbanditsasfollows:eachtimeacombinatorialstrategy,e.g.,agroupofarms,ischosen,andthedecisionmakerreceivesarewardresultingfromherstrategyandalsoreceivesasidebonusresultingfromthatstrategyforeacharmsneighbor.Thisismotivatedbymanyrealapplicationssuchason-linesocialnetworkswherefriendscanprovidetheirfeedbackonsharedcontent,thereforeifwepromoteaproducttoauser,wecanalsocollectfeedbackfromherfriendsonthatproduct.Tothisend,weconsidertwotypesofsidebonusinthisstudy:sideobservationandsidereward.Uponthenumberofarmspulledateachtimeslot,westudytwocases:single-playandcombinatorial-play.Consequently,thisleavesusfourscenariostoinvestigateinthepresenceofsidebonus:Single-playwithSideObservation,Combinatorial-playwithSideObservation,Single-playwithSideReward,andCombinatorial-playwithSideReward.Foreachcase,wepresentandanalyzeaseriesofzeroregretpoliceswheretheexpectofregretovertimeapproacheszeroastimegoestoinfinity.Extensivesimulationsvalidatetheeffectivenessofourresults.
ComputabilityofPerpetualExplorationinHighlyDynamicRings MarjorieBournat(UPMCSorbonneUniversités),SwanDubois(UPMCSorbonneUniversités),FranckPetit(UPMCSorbonneUniversités)
Weconsidersystemsmadeofautonomousmobilerobotsevolvinginhighlydynamicdiscreteenvironmenti.e.,graphswhereedgesmayappearanddisappearunpredictablywithoutanyrecurrence,stability,norperiodicityassumption.Robotsareuniform(theyexecutethesamealgorithm),theyareanonymous(theyaredevoidofanyobservableID),theyhavenomeansallowingthemtocommunicatetogether,theysharenocommonsenseofdirection,andtheyhavenoglobalknowledgerelatedtothesizeoftheenvironment.However,eachofthemisendowedwithpersistentmemoryandisabletodetectwhetheritstandsaloneatitscurrentlocation.Ahighlydynamicenvironmentismodeledbyagraphsuchthatitstopologykeepscontinuouslychangingovertime.Inthispaper,weconsideronlydynamicgraphsinwhichnodesareanonymous,eachofthemisinfinitelyoftenreachablefromanyotherone,andsuchthatitsunderlyinggraph(i.e.,thestaticgraphmadeofthesamesetofnodesandthatincludesalledgesthatarepresentatleastonceovertime)formsaringofarbitrarysize.Inthiscontext,weconsiderthefundamentalproblemofperpetualexploration:eachnodeisrequiredtobeinfinitelyoftenvisitedbyarobot.Thispaperanalyzesthecomputabilityofthisproblemin(fully)synchronoussettings,i.e.,westudythedeterministicsolvabilityoftheproblemwithrespecttothenumberofrobots.Weprovidethreealgorithmsandtwoimpossibilityresultsthatcharacterize,foranyringsize,thenecessaryandsufficientnumberofrobotstoperformperpetualexplorationofhighlydynamicrings.
LocallySelf-AdjustingSkipGraphs SikderHuq(TheUniversityofIowa),SukumarGhosh(TheUniversityofIowa)
Wepresentadistributedself-adjustingalgorithmforskipgraphsthatminimizestheaverageroutingcostsbetweenarbitrarycommunicationpairsbyperformingtopologicaladaptationtothecommunicationpattern.Ouralgorithmisfullydecentralized,conformstotheCONGESTmodel(i.e.O(logn)bitmessages),andrequiresO(logn)bitsofmemoryforeachnode,wherenisthetotalnumberofnodes.Uponeachcommunicationrequest,ouralgorithmfirstestablishescommunicationbyusingthestandardskipgraphrouting,andthenlocallyandpartiallyreconstructstheskipgraphtopologytoperformtopologicaladaptation.Weproposeacomputationalmodelforsuchalgorithms,aswellasayardstick(workingsetproperty)toevaluatethem.Ourworkingsetpropertycanalsobeusedtoevaluateself-adjustingalgorithmsforothergraphclasseswheremultipletree-likesubgraphsoverlap(e.g.hypercubenetworks).Wederivealowerboundoftheamortizedroutingcostforanyalgorithmthatfollowsourmodelandservesanunknownsequenceofcommunicationrequests.Weshowthattheroutingcostofouralgorithmisatmostaconstantfactormorethantheamortizedroutingcostofanyalgorithmconformingtoourcomputationalmodel.Wealsoshowthattheexpectedtransformationcostforouralgorithmisatmostalogarithmicfactormorethantheamortizedroutingcostofanyalgorithmconformingtoourcomputationalmodel.
OnlinetoOfflineBusiness:UrbanTaxiDispatchingwithPassenger-DriverMatchingStability HuanyangZheng(TempleUniversity),JieWu(TempleUniversity)
IntheOnlinetoOffline(O2O)taxibusiness(e.g.,Uber),theinterestsofpassengers,taxidrivers,andthecompanymaynotalignwitheachother,sincetaxisdonotbelongtothecompany.Tobalancethoseinterests,thispaperstudiesthetaxidispatchproblemfortheO2Otaxibusiness.Theinterestsofpassengersandtaxidriversaremodeled.Fornon-sharingtaxidispatches(multiplepassengerrequestscannotshareataxi),astablemarriageapproachisproposed.Itcandealwithunequalnumbersofpassengerrequestsandtaxisthroughmatchingthemtodummypartners.Theexistenceofstablematchingswithdummypartnersisproved.
![Page 70: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/70.jpg)
70
Threerulesarepresentedtofindoutallpossiblestablematchings.Forsharingtaxidispatches(multiplepassengerrequestscouldshareataxi),passengerrequestsarepackedthroughsolvingamaximumsetpackingproblem.Packedpassengerrequestsareregardedasasinglerequestformatchingtaxis.Extensiverealdata-drivenexperimentsdemonstratetheperformanceofourapproach.Theproposedalgorithmshavealimitedperformancegaptotheliteratureintermsofthedispatchdelayandthepassengersatisfactory,butsignificantlyimprovestheexistingalgorithmsintermsofthetaxisatisfactory.IndexTermsTaxidispatchschedule,passengerrequests,taxidrivers,matchingstability,sharingandnon-sharing.
AnOptimizationFrameworkForOnlineRide-sharingMarkets YongzhengJia(TsinghuaUniversity),WeiXu(TsinghuaUniversity),XueLiu(McGillUniversity)
Taxiservicesandproductdeliveryservicesareinstrumentalforourmodernsociety.Thankstotheemergenceofsharingeconomy,ride-sharingservicessuchasUber,Didi,LyftandGooglesWazeRiderarebecomingmoreubiquitousandgrowintoanintegralpartofoureverydaylives.However,theefficiencyoftheseservicesareseverelylimitedbythesub-optimalandimbalancedmatchingbetweenthesupplyanddemand.Weneedageneralizedframeworkandcorrespondingefficientalgorithmstoaddresstheefficientmatching,andhenceoptimizetheperformanceofthesemarkets.Existingstudiesfortaxianddeliveryservicesareonlyapplicableinscenariosoftheone-sidedmarket.Incontrast,thisworkinvestigatesahighlygeneralizedmodelforthetaxianddeliveryservicesinthemarketeconomy(abbreviatedastaxianddeliverymarket)thatcanbewidelyusedintwo-sidedmarkets.Further,wepresentefficientonlineandofflinealgorithmsfordifferentapplications.Weverifyouralgorithmwiththeoreticalanalysisandtrace-drivensimulationsunderrealisticsettings.
FastandAccurateTrackingofPopulationDynamicsinRFIDSystems MuhammadShahzad(NorthCarolinaStateUniversity),AlexLiu(MichiganStateUniversity)
RFIDsystemshavebeenwidelydeployedforvariousapplicationssuchassupplychainmanagement,indoorlocalization,inventorycontrol,andaccesscontrol.ThispaperdealswiththefundamentalproblemofestimatingthenumberofarrivinganddepartingtagsbetweenanytwotimeinstantsindynamicallychangingRFIDtagpopulations,whichisneededinmanyapplicationssuchaswarehousemonitoringandprivacysensitiveRFIDsystems.Inthispaper,weproposeadynamictagestimationscheme,namelyDTE,thatcanachievearbitrarilyhighrequiredreliability,iscompliantwiththeC1G2standard,andworksinsingleaswellasmultiple-readerenvironment.DTEusesthestandardizedframeslottedAlohaprotocolandutilizesthenumberofslotsthatchangetheirvaluesincorrespondingAlohaframesatthetwotimeinstantstoestimatethenumberofarrivinganddepartingtags.Itiseasytodeploybecauseitneitherrequiresmodificationtotagsnortothecommunicationprotocolbetweentagsandreaders.WehaveextensivelyevaluatedandcomparedDTEwiththeonlypriorscheme,ZDE,thatcanestimatethenumberofarrivinganddepartingtags.Unfortunately,ZDEcannotachievearbitrarilyhighrequiredreliability.Incontrast,ourproposedschemealwaysachievestherequiredreliability.Forexample,foratagpopulationcontaining104tags,arequiredreliabilityof95%,andarequiredconfidenceintervalof5%,DTEtakes5.12secondstoachievetherequiredreliabilitywhereasZDEachievesareliabilityofonly66%inthesameamountoftime.
Research14:MobileandWirelessComputingSystemsIIRobustIndoorWirelessLocalizationUsingSparseRecovery WeiGong(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity)
Withthemulti-antennadesignofWiFiinterfaces,phasedarrayhasbecomeapromisingmechanismforaccurateWiFilocalization.State-of-the-artWiFi-basedsolutionsusingAoA(Angle-of-Arrival),however,faceanumberofcriticalchallenges.First,theirlocalizationaccuracydegradesdramaticallywhentheSignal-to-NoiseRatio(SNR)becomeslow.Second,theydonotfullyutilizecoherentprocessingacrossallavailabledomains.Inthispaper,wepresentROArray,aRObustArraybasedsystemthataccuratelylocalizesatargetevenwithlowSNRs.First,inthespatialdomain,ROArraycanproducesharpAoAspectrumsbyparameterizingthesteeringvectorbasedonasparsegrid.Then,toexpandintothefrequencydomain,itjointlyestimatestheToAs(Time-of-Arrival)andAoAsofallthepathsusingmulti-subcarrierOFDMmeasurements.Furthermore,throughmulti-packetfusion,ROArrayisenabledtoperformcoherentestimationacrossthespatial,frequency,andtimedomains.Suchcoherentprocessingnotonlyincreasesthevirtualaperturesize,whichenlargesthenumberofmaximumresolvablepaths,butalsoimprovesthesystemrobustnesstonoises.Ourimplementationusingoff-the-shelfWiFicardsdemonstratesthat,withlowSNRs,ROArraysignificantlyoutperformsstate-ofthe-artsolutionsinlocalizationaccuracy;whenmediumorhighSNRsarepresent,itachievescomparableaccuracy.
Max-MinFairResourceAllocationinHetNets:DistributedAlgorithmsandHybridArchitecture EhsanAryafar(PortlandStateUniversity),AlirezaKeshavarz-Haddad(ShirazUniversity),CarleeJoe-Wong(CarnegieMellonUniversity),MungChiang(PrincetonUniversity)
WestudytheresourceallocationprobleminRANlevelintegratedHetNets.ThisemergingHetNetsparadigmallowsfordynamictrafficsplittingacrossradioaccesstechnologiesforeachclient,andthenforaggregatingthetrafficinsidethenetworktoimprovetheoverallresourceutilization.Wefocusonthemaxminfairservicerateallocationacrosstheclients,andstudythepropertiesoftheoptimalsolution.Basedontheanalysis,wedesignalowcomplexitydistributedalgorithmthattriestoachievemax-minfairness.Wealsodesignahybridnetworkarchitecturethatleveragesopportunisticcentralizednetworksupervisiontoaugmentthedistributedsolution.Weanalyzetheperformanceofourproposedalgorithmsandprovetheirconvergence.Wealsoderiveconditionsunderwhichtheoutcomeisoptimal.Whentheconditionsarenotsatisfied,weprovideconstantupperandlowerboundsontheoptimalitygap.Finally,westudytheconvergencetimeofourdistributedsolutionandshowthatleveragingappropriatepoliciesinitsdesignsignificantlyreducestheconvergencetime.
OptimizationofFull-ViewBarrierCoveragewithRotatableCameraSensors XiaofengGao(ShanghaiJiaoTongUniversity),RuiYang(UniversityofIllinoisUrbana-Champaign),FanWu(ShanghaiJiaoTongUniversity),GuihaiChen(ShanghaiJiaoTongUniversity),JingguangZhou(ShanghaiJiaoTongUniversity)
Inalltheresearchesinwirelesssensornetworks,camerasareincreasinglyutilizedfortheirsurveillancecapabilities.Inthispaper,weelaboratelydiscussabouttheproblemofFull-ViewBarrierCoveragewithRotatableCameraSensors(FBR),includingweaklyandstronglyconnectedversions.FBRisproventobeNP-hardinthispaperbyreducingGroupSteinerTreeproblemtoit.Ourgoalistoreducesensornumberwhenguaranteeingthesurveillancecapabilitiesatthesametime.Correspondingly,weintroduceanovelweighedgraphstructurecalledFull-ViewBarrierGraph.Wetransformweakversionproblemintoapseudoone-dimensiononeandproposeWGraProjalgorithmwiththehelpofdynamicprogramming;instrongversionproblem,weintroducetwocentralizedalgorithms(S-Dijkstra,S-Thorup),respectivelyaimingtosavesensornumberandtoreducetimecomplexity.Moreover,werigorouslyanalyzethecorrectnessandtimecomplexityforeachalgorithm.Inaddition,themassnumberofexperimentsareconductedtovalidatetheefficiencyofallalgorithms,whichprovethatourstructuresandalgorithmscanconstructafull-viewbarrierwithfewercamerasensorscomparedwithpreviousresearches.
![Page 71: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/71.jpg)
71
CommunicationthroughSymbolSilence:TowardsFreeControlMessagesinIndoorWLANs BingFeng(UniversityofScienceandTechnologyofChina),JianqingLiu(UniversityofFlorida),ChiZhang(UniversityofScienceandTechnologyofChina),YuguangFang(UniversityofFlorida)
Efficientdesignofwirelessnetworksbenefitsfromtheexchangeofcontrolmessages.However,controlmessageitselfconsumesscarcechannelresources.Inthispaper,weproposeCoS(CommunicationthroughsymbolSilence),anovelcommunicationstrategythatconveyscontrolmessagesforfreewithoutconsumingextrachannelresources.CoSinsertssilencesymbolsindatapacketsandleveragestheintervalsbetweeninsertedsilencesymbolstoencodeinformation.Thesilencesymbolscanbelocatedbyenergydetectionatthegranularityofsymbolsandtheintervalsareinterpretedintotransmittedcontrolmessages.Basedonourkeyinsightsthatthechannelcodeisunder-utilizedincurrentwirelessnetworksandthedistributionofsymbolerrorswithinadatapacketispredictableinindoorwirelesstransmissions,thesymbolserasedbysilencesymbolsarerecoveredbythecodingredundancythatisoriginallyusedtocorrectsymbolerrors.Arateadaptationschemeisdesignedtodynamicallyadjusttherateoffreecontrolmessagesaccordingtochannelconditionssothatthetransmissionoffreecontrolmessagesdoesnotharmtheoriginaldatathroughput.WeimplementCoSonoursoftwaredefinedradioplatformtovalidatethefeasibilityofCoS.Theextensiveresultsshowthatthecontrolmessagesaredeliveredwithcloseto100%accuracyinalargeSNRrange.Inaddition,wemeasuretheachievablecapacityoffreecontrolmessagesinvariouschannelconditions.
Secureconnectivityofwirelesssensornetworksunderkeypredistributionwithon/offchannels JunZhao(CarnegieMellonUniversity)
Theq-compositekeypredistributionscheme[1]isusedprevalentlyforsecurecommunicationsinlarge-scalewirelesssensornetworks(WSNs).Priorwork[2][4]exploressecureconnectivityinWSNsemployingtheq-compositeschemeforq=1withunreliablecommunicationlinksmodeledasindependenton/offchannels.Inthispaper,weinvestigatesecureconnectivityinWSNsoperatingundertheq-compositeschemeforgeneralqandundertheon/offchannelmodel.Wepresentconditionsonhowtoscalethemodelparameterssothatthenetworkissecurelyconnectedwithhighprobabilitywhenthenumberofsensorsbecomeslarge.Theresultsaregivenintheformofzero-onelaws.Numericalexperimentsconfirmthevalidityofouranalyticalresults.
iUpdater:LowCostRSSFingerprintsUpdatingforDevice-freeLocalization LiqiongChang(NorthwestUniversity),JieXiong(SingaporeManagementUniversity),YuWang(UniversityofNorthCarolinaatCharlotte),XiaojiangChen(NorthwestUniversity),JunhaoHu(NorthwestUniversity),FangDingyi(NorthwestUniversity)
Whilemostexistingindoorlocalizationtechniquesaredevice-based,manyemergingapplicationssuchasintruderdetectionandelderlycaredrivetheneedsofdevice-freelocalization,inwhichthetargetcanbelocalizedwithoutanydeviceattached.Amongthediversetechniques,receivedsignalstrength(RSS)fingerprint-basedmethodsarepopularbecauseofthewideavailabilityofRSSreadingsinmostcommodityhardware.However,currentfingerprint-basedsystemssufferfromhighhumanlaborcosttoupdatesofthefingerprintdatabaseandlowaccuracyduetolargedegreeofRSSvariations.Inthispaper,weproposeafingerprint-baseddevice-freelocalizationsystemnamediUpdatertosignificantlyreducethelaborcostandincreasetheaccuracy.Wepresentanovelselfaugmentedregularizedsingularvaluedecomposition(RSVD)methodintegratingthesparseattributewithuniquepropertiesofthefingerprintdatabase.iUpdaterisabletoaccuratelyupdatethewholedatabasewithRSSmeasurementsatasmallnumberofreferencelocations,thusreducingthehumanlaborcost.Furthermore,iUpdaterobservesthatalthoughtheRSSreadingsvariesalot,theRSSdifferencesbetweenboththeneighboringlocationsandadjacentwirelesslinksarerelativelystable.Thisuniqueobservationisappliedtoovercometheshort-termRSSvariationstoimprovethelocalizationaccuracy.Extensiveexperimentsinthreedifferentenvironmentsover3monthsdemonstratetheeffectivenessandrobustnessofiUpdater.
Research15:SocialNetworksandCrowdsourcingInfluenceMaximizationinaManyCascadesWorldIoulianaLitou(AUEB),VanaKalogeraki(AUEB),DimitriosGunopulos(UoA)
OnlineSocialNetworks(OSNs)arewidelyutilizedinviralmarketingcampaignsexploitingtheword-of-moutheffect.VariouspropagationmodelshavebeenproposedtodescribethewaycascadesunfoldinOSNs.Basedontheexistingpropagationmodels,severalstudiesaddresstheproblemofinfluencemaximization,wheretheobjectiveistoidentifyanappropriatesubsetofuserstoinitiatethespreadofacontagion.However,existingapproachesignoreanimportantfactorinthepropagationprocess,i.e.,thecorrelationofmultiplecontagionssimultaneouslycascadinginthesocialnetworkandhowtheseaffecttheusersdecisionsregardingtheadoptionofacontagion.Althoughrecentworkslookintoeitherthecompetitionorthecomplementarityamongapairofcontagions,auniformmodelthatdescribesthepropagationofmultiplecascadeswithvaryingtypesanddegreesofcorrelationsislacking.Thisworkconstitutesthefirstattempttofillthisgap.Weformulateanovelpropagationmodel,theCorrelatedContagionsDynamicLinearThreshold(CCDLT),thatconsidersthecorrelationofmanycontagionsineithercompetitiveorcomplementarymanner.Ourproposedmodelallowsfordifferentdegreesofcompetition/complementarityamongcascades.Wefurtherconsiderthatusersmaydynamicallyswitchstatesregardingthecontagiontheypromoteduringthepropagationprocess,basedontheinfluenceoftheirneighborhoods.Wethendesignagreedyseedselectionalgorithmthatidentifiestheappropriatesubsetofuserstoparticipateinaspecificcontagioninordertomaximizeitsspreadandweformallyprovethatitapproximatesthebestsolutionataratioof1_1/e.Throughanextensiveexperimentalevaluationwedemonstratethesuperiorityofourapproachoverexistingschemes.
Expertise-AwareTruthAnalysisandTaskAllocationinMobileCrowdsourcingXiaomeiZhang(UniversityofSouthCarolinaBeaufort),YiboWu(PennsylvaniaStateUniversity),LifuHuang(RensselaerPolytechnicInstitute),HengJi(RensselaerPolytechnicInstitute),GuohongCao(PennsylvaniaStateUniversity)
Mobilecrowdsourcinghasreceivedconsiderableattentionasitenablespeopletocollectandsharelargevolumeofdatathroughtheirmobiledevices.Sincetheaccuracyofthecollecteddataisusuallyhardtoensure,researchershaveproposedtechniquestoidentifytruthfromnoisydatabyinferringandutilizingthereliabilityofusers,andallocatetaskstouserswithhigherreliability.However,theyneglectthefactthatausermayonlyhaveexpertiseonsomeproblems(insomedomains),butnotothers.Neglectingthisexpertisediversitymaycausetwoproblems:lowestimationaccuracyintruthanalysisandineffectivetaskallocation.Toaddresstheseproblems,weproposeanExpertise-awareTruthAnalysisandTaskAllocation(ETA2)approach,whichcaneffectivelyinferuserexpertiseandthenallocatetasksandestimatetruthbasedontheinferredexpertise.ETA2reliesonanovelsemanticanalysismethodtoidentifytheexpertisedomainsofthetasksanduserexpertise,anexpertiseawaretruthanalysissolutiontoestimatetruthandlearnuserexpertise,andanexpertise-awaretaskallocationmethodtomaximizethe
![Page 72: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/72.jpg)
72
probabilitythattasksareallocatedtouserswiththerightexpertisewhileensuringtheworkloaddoesnotexceedtheprocessingcapabilityateachuser.Experimentalresultsbasedontworeal-worlddatasetsdemonstratethatETA2significantlyoutperformsexistingsolutions.
MeLoDy:ALong-termDynamicQuality-awareIncentiveMechanismforCrowdsourcing HongweiWang(ShanghaiJiaoTongUniversity),SongGuo(TheHongKongPolytechnicUniversity),JiannongCao(TheHongKongPolytechnicUniversity),MinyiGuo(ShanghaiJiaoTongUniversity)
CrowdsourcingallowsrequesterstoallocatetaskstoagroupofworkersontheInternettomakeuseoftheircollectiveintelligence.Qualitycontrolisakeydesignobjectiveinincentivemechanismsforcrowdsourcingasrequestersaimatobtaininganswersofhighqualityunderagivenbudget.However,whenmeasuringworkerslong-termquality,existingmechanismseitherfailtoutilizeworkershistoricalinformation,ortreatworkersqualityasstableandignoreitstemporalcharacteristics,henceperformingpoorlyinalongrun.InthispaperweproposeMELODY,along-termdynamicqualityawareincentivemechanismforcrowdsourcing.MELODYmodelsinteractionbetweenrequestersandworkersasreverseauctionsthatruncontinuously.IneachrunofMELODY,wedesignatruthful,individualrational,budgetfeasibleandquality-awarealgorithmfortaskallocationwithpolynomial-timecomputationcomplexityandO(1)performanceratio.Moreover,takingintoconsiderationthelong-termcharacteristicsofworkersquality,weproposeanovelframeworkinMELODYforqualityinferenceandparameterslearningbasedonLinearDynamicalSystemsattheendofeachrun,whichtakesfulladvantageofworkershistoricalinformationandpredictstheirqualityaccurately.Throughextensivesimulations,wedemonstratethatMELODYoutperformsexistingworkintermsofbothqualityestimation(reducingestimationerrorby17.6%_24.2%)andsocialperformance(increasingrequestersutilityby18.2%_46.6%)inlong-termscenarios.
TheStrongLinkGraphforEnhancingSybilDefenses SuhendryEffendy(NationalUniversityofSingapore),RolandYap(NationalUniversityofSingapore)
Thesybilproblemisafundamentalproblemindistributedsystemsandonlinesocialnetworks(OSNs).Thebasicproblemisthatanattackercaneasilycreatemultipleidentitiesinadistributedoropenonlinesystem.Popularandeffectivesybildefensesareusuallybasedonpropertiesofthenetworkstructure.However,mostdefensesassumethatitishardfortheattackertomakemanyconnectionstohonestusers.However,thisassumptioncanbeinvalidinrealOSNswhichdecreasestheeffectivenessofmanysybildefenses.Weproposeagraphtransformation,thestronglinkgraph,tomitigatesuchattacksbyreducingtheeffectofalargenumberofattackedges.Ourpreliminaryexperimentsshowindeedthatwhentheattackerhasmanyattackedges,existingalgorithmssuchasSybilLimit,SybilRankandGatekeeperareineffective.Afterthestronglinkgraphisapplied,itdeletesmanyoftheattackedges,restoringtheeffectivenessofthesybildefenses.
MechanismDesignforMobileCrowdsensingwithExecutionUncertainty ZhenzheZheng(ShanghaiJiaoTongUniversity),ZhaoxiongYang(ShanghaiJiaoTongUniversity),FanWu(ShanghaiJiaoTongUniversity),GuihaiChen(ShanghaiJiaoTongUniversity)
Mobilecrowdsensinghasemergedasapromisingparadigmfordatacollectionduetoincreasinglypervasiveandpowerfulmobiledevices.Therehavebeenextensiveresearchworksthatproposeincentivemechanismsforcrowdsensing,buttheyallmaketheassumptionthatthemobileuserwillpositivelycompletetheallocatedsensingtask.Inthispaper,weconsideranewscenarioofcrowdsensingwhereausermayfailtocompletethetask.Forexample,wesupposetheusercontinuouslycollectdatawithhisdeviceinthebackground,andhecompletesthesensingtaskonlyifhepassesthroughthelocationofinterest.Duetotheusersmobilitypattern,hemaysucceedorfailinthetask.Itisanimportantissuefortheincentivemechanismtoensurefaulttoleranceforeachsensingtask.Wedesignreverseauctionstomodeltheinteractionbetweentheplatformandmobileusers,inwhichusersprobabilityofsuccessandcosttoperformthetasksareprivateinformation,andweaimtoguaranteethetaskstobecompletedwithhighprobability,whileminimizingthesocialcost.WeprovethatminimizingthesocialcostisanNP-hardproblem,andpresentmechanismsthatachievetruthfulnessandguaranteedapproximationratio.Weperformextensivesimulationstovalidatethedesirablepropertiesofourmechanisms.
TowardsScalableandDynamicSocialSensingUsingADistributedComputingFramework DanielZhang(UniversityofNotreDame),ChaoZheng(UniversityofNotreDame),DongWang(UniversityofNotreDame),DougThain(UniversityofNotreDame),XinMu(UniversityofNotreDame),GregMadey(UniversityofNotreDame),ChaoHuang(UniversityofNotreDame)
WiththerapidgrowthofonlinesocialmediaandubiquitousInternetconnectivity,socialsensinghasemergedasanewcrowdsourcingapplicationparadigmofcollectingobservations(oftencalledclaims)aboutthephysicalenvironmentfromhumansordevicesontheirbehalf.Afundamentalprobleminsocialsensingapplicationsliesineffectivelyascertainingthecorrectnessofclaimsandthereliabilityofdatasourceswithoutknowingeitherofthemapriori,whichisreferredtoastruthdiscovery.Whilesignificantprogresshasbeenmadetosolvethetruthdiscoveryproblem,someimportantchallengeshavenotbeenwelladdressedyet.First,existingtruthdiscoverysolutionsdidnotfullysolvethedynamictruthdiscoveryproblemwherethegroundtruthofclaimschangesovertime.Second,manycurrentsolutionsarenotscalabletolarge-scalesocialsensingeventsbecauseofthecentralizednatureoftheirtruthdiscoveryalgorithms.Third,theheterogeneityandunpredictabilityofthesocialsensingdatatrafficposeadditionalchallengestotheresourceallocationandsystemresponsiveness.Inthispaper,wedevelopedaScalableStreamingTruthDiscovery(SSTD)solutiontoaddresstheabovechallenges.Inparticular,wefirstdevelopedadynamictruthdiscoveryschemebasedonHiddenMarkovModels(HMM)toeffectivelyinfertheevolvingtruthofreportedclaims.WefurtherdevelopedadistributedframeworktoimplementthedynamictruthdiscoveryschemeusingWorkQueueinHTCondorsystem.WealsointegratedtheSSTDschemewithanoptimalworkloadallocationmechanismtodynamicallyallocatetheresources(e.g.,cores,memories)tothetruthdiscoverytasksbasedontheircomputationrequirements.WeevaluatedSSTDthroughrealworldsocialsensingapplicationsusingTwitterdatafeeds.Theevaluationresultsonthreereal-worlddatatraces(i.e.,BostonBombing,ParisShootingandCollegeFootball)showthattheSSTDschemeisscalableandoutperformsthestate-of-thearttruthdiscoverymethodsintermsofbotheffectivenessandefficiency.
![Page 73: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/73.jpg)
73
Industry and Experimentation Track Paper Abstracts
Industry1:CloudDataCentersandPerformancePhoenix:Constraintawareschedulingforheterogeneousdatacenters PrashanthThinakaran(PennsylvaniaStateUniversity),JashwantRajGunasekaran(PennsylvaniaStateUniversity),BikashSharma(MicrosoftCorp),MahmutKandemir(PennsylvaniaStateUniversity),ChitaDas(PennsylvaniaStateUniversity)
Today’sdatacentersareincreasinglybecomingdiverseregardingbothhardwareandsoftwarearchitecturesinordertosupportamyriadofapplications.Theseapplicationsarealsoheterogeneousintermsofjobresponsetimesandresourcerequirements(eg.,NumberofCores,GPUs,NetworkSpeed)andtheyareexpressedastaskconstraints.Constraintsareusedforensuringtaskperformanceguarantees/QualityofService(QoS)byenablingtheapplicationtoexpressitsspecificresourcerequirements.Whileseveralschedulershaverecentlybeenproposedthataimtoimproveoverallapplicationandsystemperformance,fewoftheseschedulersconsiderresourceconstraintsacrosstaskswhilemakingtheschedulingdecisions.Furthermore,latency-criticalworkloadsandshort-livedjobsthattypicallyconstituteabout90%ofthetotaljobsinadatacenterhavestrictQoSrequirements,whichcanbemitigatedbyminimizingthetaillatencythrougheffectivescheduling.Inthispaper,weproposePhoenix,aconstraint-awarehybridschedulertoaddressboththeseproblems(constraintawarenessandensuringlowtaillatency)byminimizingthejobresponsetimesatconstrainedworkersandproactivelyreorderingthetasks.WeuseanovelConstraintResourceVector(CRV)basedscheduling,whichinturnfacilitatesreorderingofthejobsinaqueuetominimizetaillatency.WehaveusedthepubliclyavailableGoogletracestoanalyzetheirconstraintcharacteristicsandhaveembeddedtheseconstraintsinClouderaandYahooclustertracesforstudyingtheimpactoftracesonsystemperformance.ExperimentswithGoogle,ClouderaandYahooclustertracesacross15,000workernodeclustershowsthatPhoeniximprovesthe99thpercentilejobresponsetimesonanaverageby1.9xacrossallthreetraceswhencomparedagainstastate-of-the-arthybridscheduler.Further,incomparisontootherdistributedschedulerlikeHawk,itimprovesthe90thand99thpercentilejobresponsetimesby4.5xand5xrespectively.
DualScalingVMsandQueries:Cost-effectiveLatencyCurtailment JuanPérez(UniversityofMelbourne),RobertBirke(IBMResearchZurich),MathiasBjörkqvist(IBMResearchZurich),LydiaY.Chen(IBMResearchZurich)
WimpyvirtualinstancesequippedwithsmallnumbersofcoresandRAMarepopularpublicandprivatecloudofferingsbecauseoftheirlowcostforhostingapplications.Thechallengeishowtorunlatency-sensitiveapplicationsusingsuchinstances,whichtradeoffperformanceforcost.Inthisstudy,weanalyticallyandexperimentallyshowthatsimultaneouslyscalingresourcesatcoarsegranularityandworkloads,i.e.,submittingmultiplequeryclonestodifferentservers,atfinegranularitycanovercometheperformancedisadvantagesofwimpyVMinstancesandachievestringentlatencytargetsthatareevenlowerthantheaverageexecutiontimesofwimpyservers.Tosuchanend,wefirstderiveaclosed-formanalysisforthelatencyunderanygivenVMprovisioningandqueryreplicationlevel,consideringcloningpoliciesthatcan(not)terminateoutstandingcloneswith(without)anoverhead.Validatedontrace-drivensimulations,ouranalysisisabletoaccuratelypredictthelatencyandefficientlysearchfortheoptimalnumberofVMsandclones.Secondly,wedevelopadualelasticscaler,DuoScale,thatdynamicallyscalesVMsandclonesaccordingtotheworkloaddynamicssoastoachievethetargetlatencyinacost-effectivemanner.TheeffectivenessofDuoScaleliesontheobservationthattheapplicationperformanceonlyscalessub-linearlywithincreasingverticalorhorizontalresourceprovisioning,i.e.,resourcesperVMornumberofVMs.WeevaluateDuoScaleagainstVM-onlyscalingstrategiesviaextensivetrace-drivensimulationsaswellasexperimentalresultsonacloudtest-bed.OurresultsshowthatDuoScaleisabletoachievethestringenttargetlatencybyusingclonesonwimpyVMswithcostsavingsupto50%,comparedtoscalingbrawnyVMsthathavebetterperformanceatahigherunitcost.
Aframeworkforenablingsecurityservicescollaborationacrossmultipledomains DanielMigault(EricssonSecurityResearch),MarcosSimplicioJunior(EscolaPolitécnica),BrunoBarros(EscolaPolitécnica),MakanPourzandi(EricssonSecurityResearch),ThiagoAlmeida(EscolaPolitécnica),EwertonAndrade(EscolaPolitécnica),TerezaCarvalho(EscolaPolitécnica)
CollaborationamongSecurityServiceFunctions(SSF)isexpectedtobecomeasessentialtoSECaaS(SECurityasaService)systemsaselasticityistoIaaS(InfrastructureasaService).Thevirtualizationopensnewerainnetworksecurityasnewsecurityappliancescanbecreatedondemandinappropriateplacesinthenetwork.Atthesametime,theincreasingsizeanddiversityofattacksmakeitnecessarytocomeupwithnewapproachesformoreefficientandmoreresilientsecuritymechanisms.Inthispaper,weproposeanewframeworkleveragingSDN(SoftwareDefinedNetworking)andSFC(ServiceFunctionChaining)toenhancethecollaborationamongdifferentSSFstomitigatelargescaleattacks.WedescribeaframeworkthatallowsSSFsfromdifferentdomainstonegotiateanddynamicallycontroltheamountofresourcesallocatedforcollaboration,inwhatwecalla“best-effort”collaborationmode.ThisSSFcollaborationframeworkcreatesadistributedmitigationsystemforhandlinglargescaleattacksinadynamicandscalablemanner.Theefficiencyandfeasibilityofthisframeworkisexperimentallyassessed,showingthatourapproachincurslowoverhead,increasestheamountoftraffictreatedbySSFsandreducesthedroppedtrafficduetothelackofresourcesfromthesecuritymechanisms.
GroupClusteringUsingInter-GroupDissimilarities DebessayFesehayeKassa(VMware),LeninSingaravelu(Google),Chien-ChiaChen(VMware),XiaoboHuang(VMware),AmitabhaBanerjee(VMware),RuijinZhou(VMware),RajeshSomasundaran(VMware)
Varioussystemshavenaturalgroupings.Forinstanceinlargescaledistributedsystem,wecanhavegroupsofvirtualand/orphysicaldevices.Asystemcanalsohavegroupsoftimeseriesdatasetscollectedatdifferenttimeintervals.Suchgroupsareusuallycharacterizedbymultidimensionalmetrics(features)set.Clusteringsuchgroupsusingtheirmultidimensionaldatasetshasvariousapplicationssuchasidentifyingdifferentperformancelevelsforanomalydetectionandloadbalancing.Traditionalalgorithmsfocusonclusteringasingletimeseriesdatasetandnotonsuchgroupswithmultidimensionalmetricsdatasets.Inthispaperwepresentthedesign,implementationandanalysisoftwosetsofgroupclusteringalgorithms.Thefirstsetiscalledone-to-manyasitgeneratesclustersofgroupsbycomparingeachgroupagainstallothergroups.Thesecondsetofalgorithmsiscalledpairwiseasitgeneratestheclustersofgroupsusingpairwisegroupdissimilaritymatrix.Bothsetsofalgorithmsfirstgenerategroupdissimilarityweightsusingmetricrankingalgorithms.Weimplementedthegroupclusteringalgorithmsbyextendingawellknownmachinelearningpackageandusingafront-endvisualizer.WevalidatedtheclusteringalgorithmsusingrealworlddatasetsontheVMwarevSANproduct.Experimentalresultsshowthat7outofthe8proposedalgorithmscangenerateexpectedclustersinatleast4outofthe6detailedexperiments.The3out
![Page 74: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/74.jpg)
74
of8proposedalgorithmscangeneratetheexpectedclustersin5outofthe6experiments.Oneofthepairwisealgorithmscangeneratetheexpectedclustersinall6ofthe6experiments.
ComprehensiveMeasurementandAnalysisoftheUser-PerceivedI/OPerformanceinaProductionLeadership-ClassStorageSystemLipengWan(OakRidgeNationalLaboratory),MatthewWolf(OakRidgeNationalLaboratory),FeiyiWang(OakRidgeNationalLaboratory),JongYoulCho(OakRidgeNationalLaboratory),GeorgeOstruchov(OakRidgeNationalLaboratory),ScottKlasky(OakRidgeNationalLaboratory)
WiththeincreaseofthescaleandintensityoftheparallelI/Oworkloadsgeneratedbythosescientificapplicationsrunningonhighperformancecomputingfacilities,understandingtheI/Odynamics,especiallytherootcauseoftheI/OperformancevariabilityanddegradationinHPCenvironment,havebecomeextremelycriticaltotheHPCcommunity.Inthispaper,werunextensiveI/Omeasuringtestsonaproductionleadership-classstoragesystemtocapturetheperformancevariabilitiesoflarge-scaleparallelI/O.AnalyzingtheseresultsanditsstatisticcorrelationrevealedsomevaluableinsightsintothecharacteristicsofthestoragesystemandtherootcauseofI/Operformancevariability.Further,weleveragethesefindingsandproposeanI/OmiddlewaredesignrefactoringwhichcanimprovetheperformanceoftheparallelI/Obyoptimizingthedatastripingandplacement.Ourpreliminaryevaluationresultsdemonstratetheproposedapproachcanreducetheaverageper-processwritelatencybyatleast80%andthemaximumper-processwritelatencybyatleast20%.
Industry2:MobileComputingandInternetofThingsLocation:Charleston1
OntheLimitsofSubsamplingofLocationTraces MudhakarSrivatsa(IBMT.J.WatsonResearchCenter),RaghuGanti(IBMT.J.WatsonResearchCenter),PrasantMohapatra(UCDavis)
Locationdatacollectionatasocietalscaleisincreasinglybecomingcommon-examplesofthisarecallanddatadetailrecordsintelecommunicationcompanies,GPSsamplescollectedbycarcompanies,andGPSsamplesfrommobiledevicesinmappingcompanies(e.g.,Google,Microsoft).Suchlargescalemobilitydatasetshaveapplicationsinurbanplanning,networkplanning,surveillance,andreal-timetrafficestimations.Thispaperaddressestheproblemofsubsamplinglocationtraceswhilepreservingtheamountofinformationpresentinsuchdatasets.Wepresentanovelsubsamplingtechniquethatisbasedonahierarchicalgeographicalencodingmechanism(geohash),thatallowsforefficientspatialclustersampling.Weanalyzethissubsamplingtechniquethroughvariousinformationtheoreticmeasurestoquantifythetotal“amount”ofinformationinadatasetfromalocationtraceperspectiveandevaluatethesemetricsinthecontextoftwolargescalemobilitydatasetsfromtelecommunicationcompanies-oneisthatofcalldetailrecordsandthesecondisthatofdatadetailrecords.Weshowthatsubsamplingdatainboththesecasesbyasmuchas75%doesnotsignificantlyreducethetotalamountofinformation,i.e.thedatasetcanbeusedsimilartotheoriginalversion.ThispaveswayforthecreationofbetterspaceandCPUefficientmodelsthatcansupportvariousapplicationsreliantoncollectivelocationtraces.
SOM-TC:Self-organizingmapforHierarchicalTrajectoryClustering PranitaDewan(IBMTJWatsonResearchCenter),RaghuGanti(IBMTJWatsonResearchCenter),MudhakarSrivatsa(IBMTJWatsonResearchCenter)
Trajectoryclusteringtechniqueshelpdiscoverinterestinginsightsfrommovingobjectdata,includingcommonroutesforpeopleandvehicles,anomaloussub-trajectories,etc.Existingtrajectoryclusteringtechniquesfailtotakeintoaccounttheuncertaintypresentinlocationdata.Inthispaper,weinvestigatetheproblemofclusteringtrajectorydataandproposeanovelalgorithmforclusteringsimilarfullandsub-trajectoriestogetherwhilemodelinguncertaintyinthisdata.Wedescribethenecessarypre-processingtechniquesforclusteringtrajectorydata,namelytechniquestodiscretizerawlocationdatausingPossibleWorldsemanticstocapturetheinherentuncertaintyinlocationdata,andtosegmentfulltrajectoriesintomeaningfulsub-trajectories.Asabaseline,weextendthewellknownK-meansalgorithmtoclustertrajectorydata.Wethendescribeandevaluateanewtrajectoryclusteringalgorithm,SOM-TC(Self-OrganizingMapBasedTrajectoryClustering),thatisinspiredfromtheself-organizingmaptechniqueandisatleast4xfasterthanthebaselineK-meansandcurrentdensitybasedclusteringapproaches.
ProcessingEncryptedandCompressedTime-SeriesData MatúšHarvan(EnovosLuxembourgS.A.),SamuelKimoto(OpenSystems),ThomasLocher(ABBCorporateResearch),Yvonne-AnnePignolet(ABBCorporateResearch),JohannesSchneider(UniversityofLiechtenstein)
Numerousapplications,e.g.,intheindustrialsector,producelargeamountsoftime-seriesdata,whichmustbestoredandmadeavailablefordistributedprocessing.Whileoutsourcingdatastorageandprocessingtothird-partyserviceprovidersoffersmanybenefits,itraisesdataprivacyissues.Inlightofthisproblem,techniqueshavebeenproposedtoshareonlyencrypteddatawiththeremoteserviceprovider,yetthecapabilitytorunmeaningfulqueriesoverthedataispreserved.However,timeseriesdataistypicallycompressedattheservertosavespace,whichisnoteasilypossiblewhendealingwithencrypteddata.Moreover,datamustbecompressedinsuchawaythatqueriescanstillbeexecutedefficiently.Asafirststepinthisdirection,wepresentanapproachthatpreservesdataprivacy,enablescompressionattheserver,andsupportsqueryingofthestoreddata.Ourevaluationusingrealworldtime-seriesdatashowsthatourcompressionmechanismcanreducetherequiredspacedrastically.Moreover,themedianrunningtimeofallconsideredqueriesincreasesmarginally,implyingthatcompressioncanbeintroducedwithoutsacrificingperformanceofqueryexecution.
CalvinConstrained-AFrameworkforIoTApplicationsinHeterogeneousEnvironments AmardeepMehta(UmeåUniversity),RamiBaddour(UniversitádellaSvizzeraitaliana),FredrikSvensson(EricssonResearch),HaraldGustafsson(EricssonResearch),ErikElmroth(UmeåUniversity)
CalvinisanIoTframeworkforapplicationdevelopment,deploymentandexecutioninheterogeneousenvironments,thatincludesclouds,edgeresources,andembeddedorconstrainedresources.InsideCalvin,allthedistributedresourcesareviewedasoneenvironmentbytheapplication.Theframeworkprovidesmulti-tenancyandsimplifiesdevelopmentofIoTapplications,whicharerepresentedusingadataflowofapplicationcomponents(namedactors)andtheircommunication.TheideabehindCalvinposessimilaritywiththeserverlessarchitectureandcanbeseenasActorasaServiceinsteadofFunctionasaService.ThismakesCalvinverypowerfulasitdoesnotonlyscaleactorsquicklybutalsoprovidesaneasyactormigrationcapability.Inthiswork,weproposeCalvinConstrained,anextensiontotheCalvinframeworktocoverresource-constraineddevices.Duetolimitedmemoryandprocessingpowerofembeddeddevices,theconstrainedsideoftheframeworkcanonlysupportalimitedsubsetoftheCalvinfeatures.ThecurrentimplementationofCalvinConstrainedsupportsactorsimplementedinC
![Page 75: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/75.jpg)
75
aswellasPython,wherethesupportforPythonactorsisenabledbyusingMicroPythonasastaticallyallocatedlibrary,bythisweenabletheautomaticmanagementofstatevariablesandenhancecodere-usability.Aswouldbeexpected,Python-codedactorsdemandmoreresourcesoverC-codedones.Weshowthattheextraresourcesneededaremanageableoncurrentoff-the-shelvemicro-controller-equippeddeviceswhenusingtheCalvinframework.
![Page 76: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/76.jpg)
76
Applications and Experiences Track Paper Abstracts
Application1:Security,Privacy,TrustinDistributedSystemsPrivacyPreservingUser-basedRecommenderSystem ShahriarBadsha(RMITUniversity),XunYi(RMITUniversity),IbrahimKhalil(RMITUniversity),ElisaBertino(PurdueUniversity)
Withtherapiddevelopmentofthesocialnetworks,CollaborativeFiltering(CF)-basedrecommendersystemshavebeenincreasinglyprevalentandbecomewidelyacceptedbyusers.TheCF-basedtechniquesgeneraterecommendationsbycollectingprivacysensitivedatafromusers.Usually,theusersaresensitivetodisclosureofpersonalinformationand,consequently,thereareunavoidablesecurityconcernssinceprivateinformationcanbeeasilymisusedbymaliciousthirdparties.Inordertoprotectagainstbreachesofpersonalinformation,itisnecessarytoobfuscateuserinformationbymeansofanefficientencryptiontechniquewhilesimultaneouslygeneratingtherecommendationbymakingtrueinformationinaccessibletoserviceproviders.Therefore,weproposeaprivacypreservinguser-basedCFtechniquebasedonhomomorphicencryption,whichiscapableofdeterminingsimilaritiesamongusersfollowedbygeneratingrecommendationswithoutrevealinganyprivateinformation.Weintroducedifferentsemi-honestpartiestopreserveprivacyandtocarryoutintermediatecomputationsforgeneratingrecommendations.Weimplementourmethodonpubliclyavailabledatasetsandshowthatourmethodispracticalaswellasachieveshighlevelofsecurityforuserswithoutcompromisingtherecommendationaccuracy.
PrivacyPreservingOptimizationofParticipatorySensinginMobileCloudComputing YeYan(OaklandUniversity),DongHan(OaklandUniversity),TaoShu(AuburnUniversity)
Withtherapidgrowthofmobilecloudcomputing,participatorysensingemergesasanewparadigmtoexploreourphysicalworldatanunprecedentedfinegranularitybyrecruitingthepervasivesensor-enabledsmartphones.Whileextensiveoptimizationhasbeenperformedintheliteraturetocoordinatethesensingactivityofthecloud-basedsensingserver(orplatform)andtheparticipatingsmartphonessoastomaximizetheefficiencyofparticipatorysensing,theprivacyissueintheoptimizationhasbeenlargelyoverlooked.Inthispaper,weproposeanovelprivacy-preservingoptimizationframeworkthatallowsboththecloud-basedplatformandmobileuserstosharedatafortheformulationandsolutionoftheoptimization,butwithoutrevealingsensitiveinformationthatmayleadtoprivacyleakageofeachother.Ourmethodisbuiltuponaprivacy-preservingversionofthewell-knownNP-hardweightedset-coverageproblem.Toaccommodateprivacyrequirementsinthisframework,oursolutionusesamodifiedbloomfilteralongwithaDiffie-Hellman-typeexchangeprotocolamongallparticipantsfordataaggregation,sharing,andpresentation.Throughextensivesimulationweevaluatetheprivacystrengthoftheproposedapproachandalsoverifyitseffectivenessandlowoverhead.
SPHINX:APasswordStorethatPerfectlyHidesPasswordsfromItself MalihehShirvanian(UniversityofAlabamaatBirmingham),StanislawJarecki(UniversityofCaliforniaatIrvine),HugoKrawczyk(IBMResearch),NiteshSaxena(UniversityofAlabamaatBirmingham)
Passwordmanagers(akastoresorvaults)allowausertostoreandretrieve(usuallyhigh-entropy)passwordsforhermultiplepassword-protectedservicesbyinteractingwitha“device”servingtheroleofthemanager(e.g.,asmartphoneoranonlinethird-partyservice)onthebasisofasinglememorable(low-entropy)masterpassword.Existingpasswordmanagersworkwelltodefeatofflinedictionaryattacksuponwebservicecompromise,assumingtheuseofhigh-entropypasswordsisenforced.However,theyarevulnerabletoleakageofallpasswordsintheeventthedeviceiscompromised,duetotheneedtostorethepasswordsencryptedunderthemasterpasswordand/ortheneedtoinputthemasterpasswordtothedevice(asinsmartphonemanagers).Evidenceexiststhatpasswordmanagerscanbeattractiveattacktargets.Inthispaper,weintroduceanovelapproachtopasswordmanagement,calledSPHINX,whichremainssecureevenwhenthepasswordmanageritselfhasbeencompromised.InSPHINX,theinformationstoredonthedeviceisinformationtheoreticallyindependentoftheuser’smasterpassword—anattackerbreakingintothedevicelearnsnoinformationaboutthemasterpasswordortheuser’ssite-specificpasswords.Moreover,anattackerwithfullcontrolofthedevice,evenatthetimetheuserinteractswithit,learnsnothingaboutthemasterpassword—thepasswordisnotenteredintothedeviceinplaintextformorinanyotherwaythatmayleakinformationonit.Unlikeexistingmanagers,SPHINXproducesstrictlyhigh-entropypasswordsandmakesitcompulsoryfortheuserstoregistertheserandomizedpasswordswiththewebservices,hencefullydefeatingofflinedictionaryattackuponservicecompromise.ThedesignandsecurityofSPHINXisbasedonthedevice-enhancedPAKEmodelofJareckietal.thatprovidesthetheoreticalbasisforthisconstructionandisbackedbyrigorouscryptographicproofsofsecurity.WhileSPHINXissuitablefordifferentdeviceandonlineplatforms,inthispaper,wereportonitsconcreteinstantiationonsmartphonesgiventheirpopularityandtrustworthinessaspasswordmanagers(oreventwo-factorauthentication).Wepresentthedesign,implementationandperformanceevaluationofSPHINX,offeringprototypebrowserplugins,smartphoneappsandtransparentdevice-clientcommunication.Basedonourinspectionanalysis,theoveralluserexperienceofSPHINXimprovesuponcurrentmanagers.Wealsoreportonalab-basedusabilitystudyofSPHINX,whichindicatesthatusers’perceptionofSPHINXsecurityandusabilityishighandsatisfactorywhencomparedtoregularpassword-basedauthentication.Finally,wediscusshowSPHINXmaybeextendedtoanonlineserviceforthepurposeofback-uporasanindependentpasswordmanager.
WhenSmartTVMeetsCRN:Privacy-preservingFine-grainedSpectrumAccess ChaowenGuan(StateUniversityofNewYorkatBuffalo),AzizMohaisen(StateUniversityofNewYorkatBuffalo),ZhiSun(StateUniversityofNewYorkatBuffalo),LuSu(StateUniversityofNewYorkatBuffalo),KuiRen(StateUniversityofNewYorkatBuffalo),YalingYang(VirginiaTech)
DynamicspectrumsharingtechniquesappliedintheUHFTVbandhavebeendevelopedtoallowsecondaryWiFitransmissioninareaswithactiveTVusers.Thistechniqueofdynamicallycontrollingtheexclusionzoneenablesvastlyincreasingsecondaryspectrumre-use,comparedtothe``TVwhitespace''modelwhereTVtransmittersdeterminetheexclusionzoneandonly"idle"channelscanbere-purposed.However,incurrentsuchdynamicspectrumsharingsystems,thesensitiveoperationparametersofbothprimaryTVusers(PUs)andsecondaryusers(SUs)needtobesharedwiththespectrumdatabasecontroller(SDC)forthepurposeofrealizingefficientspectrumallocation.SincesuchSDCserverisnotnecessarilyoperatedbyatrustedthirdparty,thosecurrentsystemsmightcauseessentialthreatenstotheprivacyrequirementfrombothPUsandSUs.Toaddressthisprivacyissue,thispaperproposesaprivacy-preservingspectrumsharingsystembetweenPUsandSUs,whichrealizesthespectrumallocationdecisionprocessusingefficientmulti-partycomputation(MPC)technique.Inthisdesign,theSDConlyperformssecurecomputationoverencryptedinputfromPUsandSUssuchthatnoneofthePUorSUoperationparameterswillberevealedtoSDC.TheevaluationofitsperformanceillustratesthatourproposedsystembasedonefficientMPCtechniquescanperformdynamicspectrumallocationprocessbetweenPUsandSUsefficientlywhilepreservingusers'privacy.
![Page 77: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/77.jpg)
77
RevisitingSecurityRisksofAsymmetricScalarProductPreservingEncryptionandItsVariants WeipengLin(SimonFraserUniveristy),KeWang(SimonFraserUniversity),ZhilinZhang(SimonFraserUniversity),HongChen(RenminUniversityofChina)
Cloudcomputinghasemergedasacompellingvisionformanagingdataanddeliveringqueryansweringcapabilityovertheinternet.Thisnewwayofcomputingalsoposesarealriskofdisclosingconfidentialinformationtothecloud.Searchableencryptionaddressesthisissuebyallowingthecloudtocomputetheanswertoaquerybasedontheciphertextsofdataandqueries.Thankstoitsinnerproductpreservationproperty,theasymmetricscalar-product-preservingencryption(ASPE)hasbeenadoptedandenhancedinagrowingnumberofworkstoperformavarietyofqueriesandtasksinthecloudcomputingsetting.However,thesecuritypropertyofASPEanditsenhancedschemeshasnotbeenstudiedcarefully.Inthispaper,weshowacompletedisclosureofASPEandseveralpreviouslyunknownsecurityrisksofitsenhancedschemes.Meanwhile,efficientalgorithmsareproposedtolearntheplaintextofdataandqueriesencryptedbytheseschemeswithlittleornoknowledgebeyondtheciphertexts.Wedemonstratetheserisksonrealdatasets.
AnAdversary-CentricBehaviorModelingofDDoSAttacks AnWang(GeorgeMasonUniversity),AzizMohaisen(SUNYBuffalo),SongqingChen(GeorgeMasonUniversity)
DistributedDenialofService(DDoS)attacksaresomeofthemostpersistentthreatsontheInternettoday.TheevolutionofDDoSattackscallsforanin-depthanalysisofthoseattacks.Abetterunderstandingoftheattackers’behaviorcanprovideinsightstounveilpatternsandstrategiesutilizedbyattackers.Thepriorartontheattackers’behavioranalysisoftenfallsintwoaspects:itassumesthatadversariesarestatic,andmakescertainsimplifyingassumptionsontheirbehavior,whichoftenarenotsupportedbyrealattackdata.Inthispaper,wetakeadata-drivenapproachtodesigningandvalidatingthreeDDoSattackmodelsfromtemporal(e.g.,attackmagnitudes),spatial(e.g.,attackerorigin),andpatiotemporal(e.g.,attackinter-launchingtime)perspectives.Wedesignthesemodelsbasedontheanalysisoftracesconsistingofmorethan50,000verifiedDDoSattacksfromindustrialmitigationoperations.EachmodelisalsovalidatedbytestingitseffectivenessinaccuratelypredictingfutureDDoSattacks.ComparisonsagainstsimpleintuitivemodelsfurthershowthatourmodelscanmoreaccuratelycapturetheessentialfeaturesofDDoSattacks.
Application2:SocialNetworksandCrowdsourcingAnti-MaliciousCrowdsourcingUsingtheZero-DeterminantStrategy QinHu(BeijingNormalUniversity),ShenglingWang(BeijingNormalUniversity),LiranMa(TexasChristianUniversity),RongfangBie(BeijingNormalUniversity),XiuzhenCheng(GeorgeWashingtonUniversity)
Crowdsourcingisapromisingparadigmtoaccomplishacomplextaskviaelicitingservicesfromalargegroupofcontributors.However,recentobservationsindicatethatthesuccessofcrowdsourcingisbeingthreatenedbythemaliciousbehaviorsofthecontributors.Inthispaper,weanalyzethemaliciousattackproblemusinganiteratedprisoner’sdilemma(IPD)gameandproposeazero-determinant(ZD)strategybasedschemebyrewardingaworker’scooperationorpenalizingthedefectionforenticinghisfinalcooperation.Boththeoreticalanalysisandsimulationstudyindicatethattheproposedalgorithmhastwoattractivecharacteristics:1)therequestorcanincentivizetheworkertokeeponcooperatingbyonlyincreasingtheshort-termpayment;and2)theproposedalgorithmisfair,sotherequestorcannotarbitrarilypenalizeaninnocentworkertoincreaseherpayoffeventhoughshecandominatethegame.Tothebestofourknowledge,wearethefirsttousetheZDstrategytostimulatebothplayerstocooperateinanIPDgame.Moreover,ourproposedalgorithmisnotrestrictedtosolvetheproblemofthemaliciouscrowdsourcing-itcanbeemployedtotackleanyproblemthatcanbeformulatedbyanIPDgame.
JPR:ExploringJointPartitioningandReplicationforTrafficMinimizationinOnlineSocialNetworks JingyaZhou(SoochowUniversity),JianxiFan(SoochowUniversity)
Ascalablestoragesystembecomesmoreimportanttodayforonlinesocialnetworks(OSNs)asthevolumeofuserdataincreasesrapidly.Key-valuestoreusesconsistenthashingtosavedatainadistributedmanner.Asadefactostandard,ithasbeenwidelyusedinproductionenvironmentsofmanyOSNs.However,therandomnatureofhashingalwaysleadstohighinter-servertraffic.Recently,partitioningandreplicationarerespectivelyproposedinmanyexistingworkswheretheformeraimstominimizetheinter-serverreadtrafficandthelatteraimstooptimizetheinter-serverwritetraffic.Nevertheless,theseparatedmannersofoptimizationcannotefficientlyreducethetraffic.Becausetheinter-serverreadtrafficischangedduringreplication.Inthispaper,wesuggestthatperformingpartitioningandreplicationsimultaneouslycouldprovideprobabilitytofurtheroptimizetraffic.Thenweformulatetheproblemasarevisedgraphpartitioningwithoverlaps,sinceoverlapspartitioningnaturallycorrespondstoreplication.Tosolvetheproblem,weproposeaJointPartitioningandReplication(JPR)scheme.ThroughextensiveexperimentswitharealworldFacebooktrace,weevaluatethatJPRsignificantlyreducesinter-servertrafficwithslightlysacrificingstoragecostcomparedtohashing,andpreservesagoodloadbalancingacrossserversaswell.
OptimizingSourceSelectioninSocialSensinginthePresenceofInfluenceGraphs HuajieShao(UIUC),ShiguangWang(UIUC),ShenLi(UIUC),ShuochaoYao(UIUC),YiranZhao(UIUC),MdTanvirAlAmin(UIUC),TarekAbdelzaher(UIUC),LanceKaplan(UIUC)
Thispaperaddressestheproblemofchoosingtherightsourcestosolicitdatafrominsensingapplicationsinvolvingbroadcastchannels,suchasthosecrowdsensingapplicationswheresourcessharetheirobservationsonsocialmedia.Thegoalistoselectsourcessuchthatexpectedfusionerrorisminimized.Weassumethatsolicitingdatafromasourceincursacostandthatthecostbudgetislimited.Contrarytootherformulationsofthisproblem,wefocusonthecasewheresomesourcesinfluenceothers.Hence,askingasourcetomakeaclaimaffectsthebehaviorofothersourcesaswell,accordingtoaninfluencemodel.Thepapermakestwocontributions.First,wedevelopananalyticmodelforestimatingexpectedfusionerror,givenaparticularinfluencegraphandsolutiontothesourceselectionproblem.Second,weusethatmodeltosearchforasolutionthatminimizesexpectedfusionerror,formulatingitasazero-oneintegernon-linearprogramming(INLP)problem.Toscaletheapproach,thepaperfurtherproposesanovelreliability-basedpruningheuristic(RPH)andasimilarity-basedlossyestimation(SLE)algorithmthatsignificantlyreducethecomplexityoftheINLPalgorithmatthecostofamodestapproximation.Theanalyticallycomputedexpectedfusionerrorisvalidatedusingbothsimulationsandreal-worlddatafromTwitter,demonstratingagoodmatchbetweenanalyticpredictionsandempiricalmeasurements.Itisalsoshownthatourmethodoutperformsbaselinesintermsofresultingfusionerror.
DynamicContractDesignforHeterogenousWorkersinCrowdsourcingforQualityControl ChenxiQiu(PennsylvaniaStateUniversity),AnnaSquicciarini(PennsylvaniaStateUniversity),SarahRajtmajer(PennsylvaniaStateUniversity),JamesCaverlee(TexasA&MUniversity)
![Page 78: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/78.jpg)
78
Crowdsourcingsitesheavilyrelyonpaidworkerstoensurecompletionoftasks.Yet,designingapricingstrategiesabletoincentivizeusers’qualityandretentionisnontrivial.Existingpaymentstrategieseithersimplysetafixedpaymentpertaskwithoutconsideringchangesinworkers’behaviors,orruleoutpoorqualityresponsesandworkersbasedoncoarsecriteria.Hence,taskrequestersmaybeinvestingsignificantlyinworkthatisinaccurateorevenmisleading.Inthispaper,wedesignadynamiccontracttoincentivizehigh-qualitywork.Ourproposedapproachoffersatheoreticallyprovenalgorithmtocalculatethecontractforeachworkerinacost-efficientmanner.Incontrasttoexistingwork,ourcontractdesignisnotonlyadaptivetochangesinworkers’behavior,butalsoadjustspricingpolicyinthepresenceofmaliciousbehavior.BoththeoreticalandexperimentalanalysisoverrealAmazonreviewtracesshowthatourcontractdesigncanachieveanearoptimalsolution.Furthermore,experimentalresultsdemonstratethatourcontractdesign1)canpromotehigh-qualityworkandpreventmaliciousbehavior,and2)outperformstheintuitivestrategyofexcludingallmaliciousworkersintermsoftherequester’sutility.
JointRequestBalancingandContentAggregationinCrowdsourcedCDN MingMa(TsinghuaUniversity),ZhiWang(TsinghuaUniversity),KunYi(TsinghuaUniversity),JiangchuanLiu(SouthChinaAgriculturalUniversity),LifengSun(TsinghuaUniversity)
RecentyearshavewitnessedanewcontentdeliveryparadigmnamedcrowdsourcedCDN,inwhichdevicesdeployedatedgenetworkcanprefetchcontentsandprovidecontentdeliveryservice.CrowdsourcedCDNoffershigh-qualityexperiencetoend-usersbyreducingtheircontentaccesslatencyandalleviatestheloadofnetworkbackbonebymakinguseofnetworkandstorageresourcesatmillionsofedgedevices.Insuchparadigm,redirectingcontentrequeststoproperdevicesiscriticalforuserexperience.TheuniquenessofrequestredirectioninsuchcrowdsourcedCDNliesthat:ononehand,thebandwidthcapacityofthecrowdsourcedCDNdevicesislimit,hencedeviceslocatedatacrowdedplacecanbeeasilyoverwhelmedwhenservingnearbyuserrequests;ontheotherhand,contentsrequestedinonedevicecanbesignificantlydifferentfromanotherone,makingrequestredirectionstrategiesusedinconventionalCDNswhichonlyaimtobalancerequestloadsineffective.Inthispaper,weexplorerequestredirectionstrategiesthattakebothworkloadbalanceofdevicesandcontentrequestedbyusersintoconsideration.Ourcontributionsareasfollows.First,weconductmeasurementstudies,coving1.8Muserswatching0.4Mvideos,tounderstandrequestpatternsincrowdsourcedCDN.Weobservethattheloadsofnearbydevicescanbeverydifferentandthecontentsrequestedatnearbydevicescanalsobesignificantlydifferent.Theseobservationsleadtoourdesignforrequestbalancingatnearbydevices.Second,weformulatetherequestredirectionproblembytakingboththecontentaccesslatencyandthecontentreplicationcostintoconsideration,andproposearequestbalancingandcontentaggregationsolution.Finally,weevaluatetheperformanceofourdesignusingtrace-drivensimulations,andobserveourschemeoutperformsthetraditionalstrategyintermsofmanymetrics,e.g.,weobserveacontentaccesslatencyreductionby50%overtraditionalmechanismssuchastheNearest/Randomrequestroutingscheme.
Shrink:ABreastCancerRiskAssessmentModelBasedonMedicalSocialNetwork AliLi(UniversityofScienceandTechnologyBeijing),RuiWang(UniversityofScienceandTechnologyBeijing),LeiXu(UniversityofScienceandTechnologyBeijing)
Breastcancerriskassessmentmodelcanassesswhetherapeopleisatahighriskofdevelopingbreastcancerdiseaseornotandconfirmabreastcancerhigh-riskgroup.Becausetheetiologyofbreastcancerdiseaseisdifferentindifferentcountryandregion,theexistingriskassessmentmodelisonlyadaptivetocertaincountriesandregions.Andtheparametersofthesemodelsarefixed,sothesemodelshavepoorgenerality.Aimingattheseproblems,thepaperputsforwardanewbreastcancerriskassessmentmodelnamedasShrink.Usingtheideaofsocialnetwork,Shrinkconstructsamedicalsocialnetworktoshowthesimilarityamongpeople,andusesgroupdivisionalgorithmtodividethenetworkintobreastcancerhigh-riskgroupandlow-riskgroup.Theparametersofthismodelcanbesetaccordingtotheneedsofthebreastcensus,andtheseparameterscanbedirectlyacquiredthroughquestionnaire,thereforeShrinkhasgoodgenerality.Moreover,undertheuncertainclassificationstandard,Shrinkadoptsanewclassificationmethodtodiscoverbreastcancerhigh-riskgroup.Basedontherealdatafromquestionnaires,wemakeexperimentsinMatlab,andobtaintheevaluationindexofthemodel.TheexperimentprovesthatthemodelitselfhasgoodevaluationresultandisbetterthanclassicGailmodel.
Application3:InternetofThings,SmartCities,andCyber-PhysicalSystemsOpportunisticEnergySharingBetweenPowerGridandElectricVehicles:AGameTheory-BasedPricingPolicy AnkurSarker(UniversityofVirginia),ZhuozhaoLi(UniversityofVirginia),WilliamKolodzey(ClemsonUniversity),HaiyingShen(UniversityofVirginia)
Electricvehicles(EVs)havegreatpotentialtoreducedependencyonfossilfuels.TherecentsurgeinthedevelopmentofonlineEV(OLEV)willhelptoaddressthedrawbacksassociatedwithcurrentgenerationEVs,suchastheheavyandexpensivebatteries.OLEVsareintegratedwiththesmartgridofpowerinfrastructurethroughawirelesspowertransfersystem(WPT)toincreasethedrivingrangeoftheOLEV.However,theintegrationofOLEVswiththegridcreatesatremendousloadforthesmartgrid.Thedemandofapowergridchangesovertimeandthepriceofpowerisnotfixedthroughouttheday.ThereshouldbesomecongestionavoidanceandloadbalancingpolicyimplicationstoensurequalityofservicesforOLEVs.Inthispaper,first,weconductananalysistoshowtheexistenceofunpredictablepowerloadandcongestionbecauseofOLEVs.WeusetheSimulationforUrbanMObilitytoolandhourlytrafficcountsofaroadsectionoftheNewYorkCitytoanalyzetheamountofenergyOLEVscanreceiveatdifferenttimesoftheday.Then,wepresentagametheorybasedonadistributedpowerscheduleframeworktofindtheoptimalschedulebetweenOLEVsandsmartgrid.Intheproposedframework,OLEVsreceivetheamountofpowerchargingfromthesmartgridbasedonapowerpaymentfunctionwhichisupdatedusingbestresponsestrategy.Weprovethattheupdatedpowerrequestsconvergetotheoptimalpowerschedule.Inthisway,thesmartgridmaximizesthesocialwelfareofOLEVs,whichisdefinedasmixedconsiderationoftotalsatisfactionanditspowerchargingcost.Finally,weverifytheperformanceofourproposedpricingpolicyunderdifferentscenariosinasimulationstudy.
EnergyEfficientObjectDetectioninCameraSensorNetworks TuanDao(UCRiverside),KarimKhalil(UCRiverside),AmitRoy-Chowdhury(UCRiverside),SrikanthKrishnamurthy(UCRiverside),LanceKaplan(U.S.ArmyResearchLaboratory)
Awirelesscameranetworkcanprovidesituationawarenessinformation(e.g.,humansindistress)inscenariossuchasdisasterrecovery.Ifsuchcamerasensorsarebatteryoperated,sendingrawvideofeedsbacktoacentralcontrollercanbeexpensiveintermsofenergyconsumption.Further,ifallcamerasweretousetheoptimalprocessingalgorithmforobjectdecision,theymayalsoexpendunnecessaryenergy.Statedotherwise,camerasthatcapturethesameobjectsmaynotallhavetousetheoptimalalgorithmtoachieveadesiredaccuracy,andthiscansaveprocessingenergycosts.Inthispaper,ourobjectiveistodesignandimplementaframeworkthatcansupportcoordinationamongcamerastodeliverhighlyaccuratedetectionofobjectsinanenergyefficientway.Theframework,whichwecallEECS(forenergyefficientcamerasensors),estimatesthedetectionaccuracyandenergycostsincurred(boththeprocessingandcommunicationcostsaretakeninto
![Page 79: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/79.jpg)
79
account)witheachdetectionalgorithmforeachcamera,andcomesupwithachoiceofcamerasforsendinginformationpertainingtotheobjectofinterest.Thissetofcamerasandthevideoprocessingalgorithmsthattheymustuse,arechosensoastominimizetheenergyexpenditures,givenadesireddetectionaccuracy.WeimplementEECSonacameranetworkbuiltwithsmartphones,anddemonstratethatitreducestheenergyconsumptionbyupto40%whileensuringaobjectdetectionaccuracyofover86%.
DeepOpp:Context-awareMobileAccesstoSocialMediaContentonUndergroundMetroSystems DiWu(HunanUniversity&ImperialCollegeLondon),DmitriArkhipov(UniversityofCaliforniaIrvine),ThomasPrzepiorka(ImperialCollegeLondon),QiangLiu(DartmouthCollege),JulieMcCann(ImperialCollegeLondon),AmeliaRegan(UniversityofCaliforniaIrvine)
Accessingonlinesocialmediacontentonundergroundmetrosystemsisachallengeduetothefactthatpassengersoftenloseconnectivityforlargepartsoftheircommute.Astheoldestmetrosystemintheworld,theLondonundergroundrepresentsatypicaltransportationnetworkwithintermittentInternetconnectivity.Todealwithdisruptioninconnectivityalongthesub-surfaceanddeep-levelundergroundlinesontheLondonunderground,wehavedesignedacontext-awaremobilesystemcalledDeepOppthatenablesefficientofflineaccesstoonlinesocialmediabyprefetchingandcachingcontentopportunisticallywhensignalavailabilityisdetected.DeepOppcanmeasure,crowdsourceandpredictsignalcharacteristicssuchasstrength,bandwidthandlatency;itcanusethesepredictionsofmobilenetworksignaltoactivateprefetching,andthenemployanoptimizationroutinetodeterminewhichsocialcontentshouldbecachedinthesystemgivenreal-timenetworkconditionsanddevicecapacities.DeepOpphasbeenimplementedasanAndroidapplicationandtestedontheLondonUnderground;itshowssignificantimprovementoverexistingapproaches,e.g.reducingtheamountofpowerneededtoprefetchsocialmediaitemsby2.5times.WhileweusetheLondonUndergroundtotestoursystem,itisequallyapplicableinNewYork,Paris,Madrid,Shanghai,oranyotherurbanundergroundmetrosystem,orindeedinanysituationinwhichusersexperiencelongbreaksinconnectivity.
PhaseBeat:ExploitingCSIPhaseDataforVitalSignMonitoringwithCommodityWiFiDevices XuyuWang(AuburnUniversity),ChaoYang(AuburnUniversity),ShiwenMao(AuburnUniversity)
Vitalsigns,suchasrespirationandheartbeat,areusefultohealthmonitoringsincesuchsignalsprovideimportantcluesofmedicalconditions.Effectivesolutionsareneededtoprovidecontact-free,easydeployment,low-cost,andlong-termvitalsignmonitoring.Inthispaper,wepresentPhaseBeattoexploitchannelstateinformation(CSI)phasedifferencedatatomonitorbreathingandheartbeatwithcommodityWiFidevices.WeprovidearigorousanalysisoftheCSIphasedifferencedatawithrespecttoitsstabilityandperiodicity.Basedontheanalysis,wedesignandimplementthePhaseBeatsystemwithoff-the-shelfWiFidevices,andconductanextensiveexperimentalstudytovalidateitsperformance.OurexperimentalresultsdemonstratethesuperiorperformanceofPhaseBeatoverexistingapproachesinvariousindoorenvironments.
REX:RapidEnsembleClassificationSystemforLandslideDetectionusingSocialMediaAibekMusaev(UniversityofAlabama),DeWang(GeorgiaInstituteofTechnology),JiatengXie(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology)
WestudytheproblemofusingSocialMediatodetectnaturaldisasters,ofwhichweareinterestedinaspecialkind,namelylandslides.EmployinginformationfromSocialMediapresentsuniqueresearchchallenges,asthereexistsaconsiderableamountofnoiseduetomultiplemeaningsofthesearchkeywords,suchas``landslide"and``mudslide".Totacklethesechallenges,weproposeREX,arapidensembleclassificationsystemwhichcanfilteroutnoisyinformationbyimplementingtwokeyideas:(I)anewmethodforconstructingindependentclassifiersthatcanbeusedforrapidensembleclassificationofSocialMediatexts,whereeachclassifierisbuiltusingrandomizedExplicitSemanticAnalysis;and(II)aself-correctionapproachwhichtakesadvantageoftheobservationthatthemajoritylabelassignedtoSocialMediatextsbelongingtoalargeeventishighlyaccurate.WeperformexperimentsusingrealdatafromTwitterover1.5yearstoshowthatREXclassificationachieves0.98inF-measure,whichoutperformsthestandardBag-of-Wordsalgorithmbyanaverageof0.14andthestate-of-the-artWord2Vecalgorithmby0.04.Wealsoreleasetheannotateddatasetsusedintheexperimentsasacontributiontotheresearchcommunitycontaining282klabeleditems.
TowardAnIntegratedApproachtoLocalizingFailuresinCommunityWaterNetworksQingHan(UCIrvine),PhuNguyen(UCIrvine),RonaldT.Eguchi(ImageCat),Kuo-LinHsu(UCIrvine),NaliniVenkatasubramanian(UCIrvine)
Wepresentacyber-physical-humandistributedcomputingframework,AquaSCALE,forgathering,analyzingandlocalizinganomalousoperationsofincreasinglyfailure-pronecommunitywaterservices.Today,detectionofpipebreaks/leaksinwaternetworkstakeshourstodays.AquaSCALEleveragesdynamicdatafrommultipleinformationsourcesincludingIoT(InternetofThings)sensingdata,geophysicaldata,humaninput,andsimulation/modelingenginestocreateasensor-simulation-dataintegrationplatformthatcanaccuratelyandquicklyidentifyvulnerablespots.Weproposeatwo-phaseworkflowthatbeginswithrobustsimulationmethodsusingacommercialgradehydraulicsimulator-EPANET,enhancedwiththesupportforIoTsensorandpipefailuremodelings.Itgeneratesaprofileofanomalouseventsusingdiverseplug-and-playmachinelearningtechniques.Theprofilethenincorporateswithexternalobservations(NOAAweatherreportsandtwitterfeeds)torapidlyandreliablyisolatebrokenwaterpipes.Weevaluatethetwo-phasemechanismincanonicalandreal-worldwaternetworksunderdifferentfailurescenarios.Ourresultsindicatethattheproposedapproachwithofflinelearningandonlineinferencecanlocatemultiplesimultaneouspipefailuresatfinelevelofgranularity(individualpipelinelevel)withhighlevelofaccuracywithdetectiontimereducedbyordersofmagnitude(fromhours/daystominutes).
Application4:Mobile,Wireless,andEdgeComputingMobiQoR:PushingtheEnvelopeofMobileEdgeComputingviaQuality-of-ResultOptimizationYongboLi(GeorgeWashingtonUniversity),YurongChen(GeorgeWashingtonUniversity),TianLan(GeorgeWashingtonUniversity),GuruVenkataramani(GeorgeWashingtonUniversity)
Mobileedgecomputingaimsatimprovingapplicationresponsetimeandenergyefficiencybydeployingdataprocessingattheedgeofthenetwork.DuetotheproliferationofInternetofThingsandinteractiveapplications,theever-increasingdemandforlowlatencycallsfornovelapproachestofurtherpushingtheenvelopeofmobileedgecomputingbeyondexistingtaskoffloadinganddistributedprocessingmechanisms.Inthispaper,weidentifyanewtradeoffbetweenQuality-of-Result(QoR)andserviceresponsetimeinmobileedgecomputing.Ourkeyideaismotivatedbytheobservationthatagrowingsetofedgeapplicationsinvolvingmediaprocessing,machinelearning,anddataminingcantoleratesomelevelofqualitylossinthecomputedresult.ByrelaxingtheneedforhighestQoR,significantimprovementinserviceresponsetimecanbeachieved.Towardthisend,wepresentanoveloptimizationframework,MobiQoR,whichminimizesserviceresponsetimeandappenergyconsumptionbyjointlyoptimizingtheQoRofalledgenodesandtheoffloadingstrategy.TheproposedMobiQoRisprototypedusing
![Page 80: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/80.jpg)
80
Parse,anopensourcemobileback-endtool,onAndroidsmartphones.Usingrepresentativeapplicationsincludingfacerecognitionandmovierecommendation,ourevaluationwithreal-worlddatasetsshowsthatMobiQoRreducesresponsetimeandenergyconsumptionbyupto77%(infacerecognition)and189.3%(inmovierecommendation)overexistingstrategiesunderthesamelevelofQoRrelaxation.
TruthfulAuctionsforUserDataAllowanceTradinginMobileNetworks ZhongxingMing(TsinghuaUniversity),MingweiXu(TsinghuaUniversity),NingWang(SurreyUniversity),BeijeGao(TsinghuaUniversity),QiLi(TsinghuaUniversity)
Userdataallowancetradingemergesasapromisingpracticeinmobiledatanetworkssinceitcanhelpmobilenetworkstoattractmoreusers.However,todate,thereisnostudyonuserdataallowancetradinginmobilenetworks.Inthispaper,wedevelopatruthfulframeworkthatallowsuserstobidfordataallowance.Wefocusonpreventingpricecheating,guaranteeingfairness,andminimizingtradingmaintenancecostintrading.Weformulatethedatatradingprocessasadoubleauctionproblemanddevelopalgorithmstosolvetheproblem.Inparticular,weuseauniformpriceauctionbasedonacompetitiveequilibriumtodefendagainstpricecheatingandprovidefairness.Meanwhile,weleveragelinearprogrammingtominimizetradingmaintenancecost.Weconductextensivesimulationstodemonstratetheperformanceoftheproposedmechanism.Thesimulationresultsshowthatourtradingmechanismistruthfulandfair,whileincurringaminimizedmaintenancecost.
OnlineResourceAllocationforArbitraryUserMobilityinDistributedEdgeClouds LinWang(TUDarmstadt),LeiJiao(UniversityofOregon),JunLi(UniversityofOregon),MaxMühlhäuser(TUDarmstadt)
Ascloudsmovetothenetworkedgetofacilitatemobileapplications,edgecloudprovidersarefacingnewchallengesonresourceallocation.Asusersmaymoveandresourcepricesmayvaryarbitrarily,andservicedelaysareheterogeneous,resourcesinedgecloudsmustbeallocatedandadaptedcontinuouslyinordertoaccommodatesuchdynamics.Inthispaper,wefirstformulatethisproblemwithacomprehensivemodelthatcapturesthekeychallenges,thenintroduceagap-preservingtransformationoftheproblem,andproposeanovelonlinealgorithmthatoptimallysolvesaseriesofsubproblemswithacarefullydesignedlogarithmicobjective,finallyproducingfeasiblesolutionsforedgecloudresourceallocationovertime.Wefurtherproveviarigorousanalysisthatouronlinealgorithmcanprovideaparameterizedcompetitiveratio,withoutrequiringanyaprioriknowledgeoneithertheresourcepriceortheusermobility.Throughextensiveexperimentswithbothreal-worldandsyntheticdata,wefurtherconfirmtheeffectivenessoftheproposedalgorithm.Weshowthattheproposedalgorithmachievesnear-optimalresultswithanempiricalcompetitiveratioofabout1.1,reducesthetotalcostbyupto4xcomparedtostaticapproaches,andoutperformstheonlinegreedyone-shotoptimizationsbyupto70%.
LeveragingTargetk-CoverageinWirelessRechargeableSensorNetworks PengzhanZhou(StonyBrookUniversity),CongWang(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity)
Energyremainsamajorhurdleinrunningcomputation-intensivetasksonwirelesssensors.RecenteffortshavebeenmadetoemployaMobileCharger(MC)todeliverwirelesspowertosensors,whichprovidesapromisingsolutiontotheenergyproblem.MostofpreviousworksinthisareaaimatmaintainingperpetualnetworkoperationattheexpenseofhighoperatingcostofMC.Inthemeanwhile,itisobservedthatduetolowcostofwirelesssensors,theyareusuallydeployedathighdensitysothereisabundantredundancyintheircoverageinthenetwork.Forsuchnetworks,itispossibletotakeadvantageoftheredundancytoreducetheenergycost.Inthispaper,werelaxthestrictnessofperpetualoperationbyallowingsomesensorstotemporarilyrunoutofenergywhilestillmaintainingtargetk-coverageinthenetworkatlowercostofMC.Wefirstestablishatheoreticalmodeltoanalyzetheperformanceimprovementsunderthisnewstrategy.Thenweorganizesensorsintoload-balancedclustersfortargetmonitoringbyadistributedalgorithm.Next,weproposeachargingalgorithmnamedλ-GTSPChargingAlgorithmtodeterminetheoptimalnumberofsensorstobechargedineachclustertomaintaink-coverageinthenetworkandderivetherouteforMCtochargethem.Wefurthergeneralizethealgorithmtoencompassmobiletargetsaswell.OurextensivesimulationresultsdemonstratesignificantimprovementsofnetworkscalabilityandcostsavingthatMCcanextendchargingcapabilityover2-3timeswithareductionof40%ofmovingcostwithoutsacrificingthenetworkperformance.
ReducingCellularSignalingTrafficforHeartbeatMessagesviaEnergy-EfficientD2DForwarding YanqiJin(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),XiaomengYi(HuazhongUniversityofScience&Technology),MinghuaChen(TheChineseUniversityofHongKong)
MobileInstantMessaging(IM)apps,suchasWhatsAppandWeChat,frequentlysendheartbeatmessagestoremoteserverstomaintainalways-onlinestatus.Periodicheartbeatmessagesaresmallinsize,buttheirtransmissionsincurheavysignalingtraffictofrequentlyestablishandreleasecommunicationchannelsbetweenbasestationsandsmartphones,knownassignalingstorm.Meanwhile,smartphonesalsoneedtoactivatecellulardatacommunicationmodulefrequentlyfortransmittingshortheartbeatmessages,resultinginsubstantialenergyconsumption.Toaddresstheseissues,weproposeaDevice-to-Device(D2D)basedheartbeatrelayingframework,inordertoreducesignalingtrafficandenergyconsumptioninheartbeattransmission.Theframeworkselectsthesmartphonesasrelaystoopportunisticallycollectheartbeatmessagesfromnearbysmartphonesusingenergy-efficientD2Dcommunication.ThecollectedheartbeatmessagesaretransmittedtotheBSinanaggregatedmannertoreducecellularsignalingtraffic.Basedontheperiodsandtheexpirationtimeofthecollectedheartbeatmessages,theframeworkschedulesthetransmissionsofcollectedheartbeatmessagestominimizesignalingandenergyconsumptionwhilesatisfyingtimeconstrains.WeimplementandevaluateoursolutiononAndroidsmartphones.Theresultsfromreal-worldexperimentsshowthatoursolutionachievesmorethan50%signalingtrafficreductionandupto36%energysaving.
k-ProtectedRoutingProtocolinMulti-hopCognitiveRadioNetworks Chin-JungLiu(MichiganStateUniversity),LiXiao(MichiganStateUniversity)
Incognitiveradionetworks(CRNs),theestablishedcommunicationsessionsbetweensecondaryusers(SUs)maybeaffectedorevengetinterruptedbecausetheSUsneedtorelinquishthespectrumwhenthelicensedusers(PUs)appearandreclaimthespectrum/channel.OndetectingPUactivities,theSUsontheaffectedlinkseitherswitchtoanotheravailableidlespectrumusingthesamelinkortheSUsseekforanalternativepath/link.Ineitherapproach,theongoingsessionisdestinedtoexperiencedelayorevengetsinterrupted,whichisintolerabletoqualityofservice-sensitiveapplicationssuchasmultimediastreamingoraudio/videoconferencing.Inthispaper,westudytheproblemofestablishingk-protectedroutesinCRNs.Ak-protectedrouteconsistsofasetofmainlinkswithpreassignedbackupspectrumandbackuppathsandisguaranteedtosustainfromkPUappearanceswithoutbeinginterrupted.ForaCRN,wefindak-protectedrouteforeachsessionrequestandmaximizethenumberofsessionsthatcanbesupported.Weproposebothcentralizedanddistributedk-protectedroutingalgorithmsforthis
![Page 81: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/81.jpg)
81
problem.Simulationresultsshowthatourk-protectedroutingprotocoloutperformsexistingopportunisticspectrumswitchingapproachesintermsofdelayandinterruptionrate.
Application5:CloudComputingandDataCenterSystemsMulti-ResourceLoadBalancingforVirtualNetworkFunctions TaoWang(HuazhongUniversityofScience&Technology),HongXu(CityUniversityofHongKong),FangmingLiu(HuazhongUniversityofScienceandTechnology)
Middleboxesarewidelydeployedtoperformvariousnetworkfunctionstoensuresecurityandimproveperformance.TherecenttrendofNetworkFunctionVirtualization(NFV)makesiteasyforoperatorstodeploysoftwareimplementationsofthesenetworkfunctionsoncommodityservers.However,virtualnetworkfunctionsconsumedifferentamountsofresourceswhenprocessingpackets.Thusamulti-resourceloadbalancing(MRLB)mechanismisneededtoefficientlyutilizeserverresources.MRLBprobleminthecontextofNFVisfundamentallydifferentfrommulti-resourceallocationproblems,aswellastraditionalsingle-resourceloadbalancingandmulti-resourceloadbalancingproblemsintaskscheduling.Inthispaper,wetackletheMRLBprobleminNFVbyfirstproposingdominantload—theloadofthemoststressedresourceonaserver—astheloadbalancingmetric.WethenformulatetheMRLBproblemasanoptimizationtominimizethemaximumdominantloadofallNFVserversgiventhedemand.BasedonproximalJacobianADMM,weproposeanefficientalgorithmtosolvetheprobleminlargescalesettings.Throughextensivetrace-drivensimulationsandprototypeexperimentsonatestbed,weshowthatourMRLBalgorithmwithdominantloadperformssignificantlybetterandfasterthanbenchmarkingalgorithms.
Learningfromfailureacrossmultipleclusters:Atrace-drivenapproachtounderstanding,predicting,andmitigatingjobterminations NosaybaEl-Sayed(MIT),HongyuZhu(UniversityofToronto),BiancaSchroeder(UniversityofToronto)
Inlarge-scalecomputingplatforms,jobsarepronetointerruptionsandprematureterminations,limitingtheirusabilityandleadingtosignificantwasteinclusterresources.Inthispaper,wetacklethisprobleminthreesteps.First,weprovideacomprehensivestudybasedonlogdatafrommultiplelarge-scaleproductionsystemstoidentifypatternsinthebehaviourofunsuccessfuljobsacrossdifferentclustersandinvestigatepossiblerootcausesbehindjobtermination.Ourresultsrevealseveralinterestingpropertiesthatdistinguishunsuccessfuljobsfromothers,particularlyw.r.t.resourceconsumptionpatternsandjobconfigurationsettings.Secondly,wedesignamachinelearning-basedframeworkforpredictingjobandtaskterminations.Weshowthatjobfailurescanbepredictedrelativelyearlywithhighprecisionandrecall,andalsoidentifyattributesthathavestrongpredictivepowerofjobfailure.Finally,wedemonstrateinaconcreteusecasehowourpredictionframeworkcanbeusedtomitigatetheeffectofunsuccessfulexecutionusinganeffectivetask-cloningpolicythatwepropose.
RBAY:AScalableandExtensibleInformationPlaneforFederatingDistributedDatacenterResources XinChen(GeorgiaInstituteofTechnology),LitingHu(FloridaInternationalUniversity),DouglasM.Blough(GeorgiaInstituteofTechnology),MichaelA.Kozuch(IntelLabsPittsburgh),MatthewWolf(OakRidgeNationalLaboratory)
Whilemanyinstitutions,whetherindustrial,academic,orgovernmental,satisfytheircomputingneedsthroughpubliccloudproviders,manyothersstillmanagetheirownresources,oftenasgeographicallydistributeddatacenters.Sparecapacityfromthesegeographicallydistributeddatacenterscouldbeofferedtoothers,providedtherewereamechanismtodiscover,andthenrequesttheseresources.Unfortunately,singledatacenteradministratorstendnottocooperateduetoissuesofscalability,diverseadministrativepolicies,andsite-specificmonitoringinfrastructure.ThispaperdescribesRBAY,anintegratedinformationplanethatenablessecureandscalablesharingbetweengeographicallydistributeddatacenters.RBAY'skeydesignfeaturesaretwofold.First,RBAYemploysadecentralized`hierarchicalaggregationtree'structuretoseamlesslyaggregatespareresourcesfromgeographicallydistributeddatacenterstoaglobalinformationplane.Second,RBAYattachestoeachparticipatingservera`admin-customized'handler,whichfollowssite-specificpolicytoexpose,hide,add,removeresourcestoRBAY,andthusfulfillthetaskof`whichresourcetoexposetowhom,when,andhow'.Anexperimentalevaluationoneightreal-worldgeo-distributedsitesdemonstratesRBAY'srapidresponsetocompositequeries,aswellasitsextensible,scalable,andlightweightnature.
Task-awareTCPinDataCenterNetworks SenLiu(CentralSouthUniversity),JiaweiHuang(CentralSouthUniversity),YutaoZhou(CentralSouthUniversity),JianxinWang(CentralSouthUniversity),TianHe(UniversityofMinnesota)
Inmoderndatacenters,manyflow-basedandtask-basedschemeshavebeenproposedtospeedupthedatatransmissioninordertoprovidefast,reliableservicesformillionsofusers.However,existingflow-basedschemestreatallflowsinisolation,contributinglesstoorevenhurtinguserexperienceduetothestalledflows.Otherprevalenttask-basedapproaches,suchascentralizedanddecentralizedscheduling,aresophisticatedorunabletosharetaskinformation.Inthiswork,wefirstrevealthatrelinquishingbandwidthofleadingflowstothestalledoneseffectivelyreducesthetaskcompletiontime.Wefurtherpresentthedesignandimplementationofageneralsupportingschemethatsharestheflow-tardinessinformationthroughareceiver-drivencoordination.Ourschemecanbeflexiblyandwidelyintegratedwiththestate-of-the-artTCPprotocolsdesignedfordatacenters,whilemakingnomodificationonswitches.Throughthetestbedexperimentsandsimulationsoftypicaldatacenterapplications,weshowthatourschemereducesthetaskcompletiontimeby70%and50%comparedwiththeflow-basedprotocols(e.g.DCTCP,L2DCT)andtask-basedscheduling(e.g.Baraat),respectively.Moreover,ourschemealsooutperformsotherapproachesby18%to25%inprevalenttopologiesofdatacenter.
LimitationsofLoadBalancingMechanismsforN-TierSystemsinthePresenceofMillibottlenecks TaoZhu(GeorgiaInstituteofTechnology),JackLi(GeorgiaInstituteofTechnology),JoshKimball(GeorgiaInstituteofTechnology),JunheePark(IndianaUniversity),Chien-AnLai(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology),QingyangWang(LouisianaStateUniversity)
Thescalabilityofn-tiersystemsreliesoneffectiveloadbalancingtodistributeloadamongtheserversofthesametier.Wefoundthatloadbalancingmechanisms(andsomepolicies)inserversusedintypicaln-tiersystems(e.g.,ApacheandTomcat)haveissuesofinstabilitywhenverylongresponsetime(VLRT)requestsappearduetomillibottlenecks,veryshortbottlenecksthatlastonlytenstohundredsofmilliseconds.Experimentswithstandardn-tierbenchmarksshowthatduringmillibottlenecks,someloadbalancingpolicy/mechanismcombinationsmakethemistakeofsendingnewrequeststothenode(s)sufferingfrommillibottlenecks,insteadoftheidlenodesasloadbalancersaresupposedtodo.Severalofthesemistakesareduetotheimplicitassumptionsmadebyloadbalancingpoliciesand
![Page 82: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/82.jpg)
82
mechanismsonthestabilityofsystemstate.OurstudyshowsthatappropriateremediesatpolicyandmechanismlevelscanavoidthesemistakesduringmillibottlenecksandremovetheVLRTrequests,thusimprovingtheaverageresponsetimebyafactorof12.
PerformanceAnalysisofCloudComputingCentersServingParallelizableRenderingJobsUsingM/M/c/rQueuingSystems XiulinLi(ShandongUniversity),LiPan(ShandongUniversity),JiweiHuang(BeijingUniversityofPostsandTelecommunications),ShijunLiu(ShandongUniversity),YuliangShi(ShandongUniversity),CaltonPu(GeorgiaInstituteofTechnology)
Performanceanalysisiscrucialtothesuccessfuldevelopmentofcloudcomputingparadigm.Anditisespeciallyimportantforacloudcomputingcenterservingparallelizableapplicationjobs,fordeterminingaproperdegreeofparallelismcouldreducethemeanserviceresponsetimeandthusimprovetheperformanceofcloudcomputingobviously.Inthispaper,takingthecloudbasedrenderingserviceplatformasanexampleapplication,weproposeanapproximateanalyticalmodelforcloudcomputingcentersservingparallelizablejobsusingM/M/c/rqueuingsystems,bymodelingtherenderingserviceplatformasamulti-stationmulti-serversystem.Wesolvetheproposedanalyticalmodeltoobtainacompleteprobabilitydistributionofresponsetime,blockingprobabilityandotherimportantperformancemetricsforgivencloudsystemsettings.Thusthismodelcanguidecloudoperatorstodetermineapropersetting,suchasthenumberofservers,thebuffersizeandthedegreeofparallelism,forachievingspecificperformancelevels.Throughextensivesimulationsbasedonbothsyntheticdataandreal-worldworkloadtraces,weshowthatourproposedanalyticalmodelcanprovideapproximateperformancepredictionresultsforcloudcomputingcentersservingparallelizablejobs,eventhosejobarrivalsfollowdifferentdistributions.
Application6:BigDataSystemsandDistributedDataManagementandAnalyticsEvaluationofDeepLearningFrameworksoverDifferentHPCArchitectures ShayanShams(LouisianaStateUniversity),RichardPlatania(LouisianaStateuniversity),KisungLee(LouisianaStateUniversity),Seung-JongPark(LouisianaStateUniversity)
Recentadvancesindeeplearninghaveenabledresearchersacrossmanydisciplinestouncovernewinsightsaboutlargedatasets.Deepneuralnetworkshaveshownapplicabilitytoimage,time-series,textual,andotherdata,allofwhichareavailableinaplethoraofresearchfields.However,theircomputationalcomplexityandlargememoryoverheadrequiresadvancedsoftwareandhardwaretechnologiestotrainneuralnetworksinareasonableamountoftime.Tomakethispossible,therehasbeenaninfluxindevelopmentofdeeplearningsoftwarethataimtoleverageadvancedhardwareresources.Inordertobetterunderstandtheperformanceimplicationsofdeeplearningframeworksoverthesedifferentresources,weanalyzetheperformanceofthreedifferentframeworks,Caffe,TensorFlow,andApacheSINGA,overseveralhardwareenvironments.Thisincludesscalingupandoutwithsingle-andmulti-nodesetupsusingdifferentCPUandGPUtechnologies.Notably,weinvestigatetheperformancecharacteristicsofNVIDIA'sstate-of-the-arthardwaretechnology,NVLink,andalsoIntel'sKnightsLanding,themostadvancedIntelproductfordeeplearning,withrespecttotrainingtimeandutilization.Toourbestknowledge,thisisthefirstworkconcerningdeeplearningbench-markingwithNVLinkandKnightsLanding.Throughtheseexperiments,weprovideanalysisoftheframeworks'performanceoverdifferenthardwareenvironmentsintermsofspeedandscaling.Asaresultofthiswork,betterinsightisgiventowardsbothusinganddevelopingdeeplearningtoolsthatcatertocurrentandupcominghardwaretechnologies.
OnAchievingEfficientDataTransferforGraphProcessinginGeo-DistributedDatacenters AmelieChiZhou(InriaRennes),ShadiIbrahim(InriaRennes),BingshengHe(NationalUniversityofSingapore)
Graphpartitioningisimportantforoptimizingtheperformanceandcommunicationcostoflargegraphprocessingjobs.Recently,manygraphapplicationssuchassocialnetworksstoretheirdataongeo-distributeddatacenters(DCs)toprovideservicesworldwidewithlowlatency.Thisraisesnewchallengestoexistinggraphpartitioningmethods,duetothecostlyWideAreaNetwork(WAN)usageandthemulti-levelsofnetworkheterogeneitiesingeo-distributedDCs.Inthispaper,weproposeageo-awaregraphpartitioningmethodnamedG-Cut,whichaimsatminimizingtheinter-DCdatatransfertimeofgraphprocessingjobsingeo-distributedDCswhilesatisfyingtheWANusagebudget.G-CutadoptstwonoveloptimizationphaseswhichaddressthetwochallengesinWANusageandnetworkheterogeneitiesseparately.G-Cutcanbealsoappliedtopartitiondynamicgraphsthankstoitslight-weightruntimeoverhead.WeevaluatetheeffectivenessandefficiencyofG-Cutusingrealworldgraphswithbothrealgeo-distributedDCsandsimulations.EvaluationresultsshowthatG-Cutcanreducetheinter-DCdatatransfertimebyupto58%andreducetheWANusagebyupto70%comparedtostate-of-the-artgraphpartitioningmethodswithalowruntimeoverhead.
GBooster:TowardsAccelerationofGPU-intensiveMobileApplications ElliottWen(VictoriaUniversityofWellington),BryanNg(VictoriaUniversityofWellington),WinstonSeah(VictoriaUniversityofWellington),XueLiu(McGillUniversity),JiannongCao(TheHongKongPolytechnicUniversity),XuefengLiu(HuangzhongUniversityofScienceandTechnology)
TheperformanceofGPUsonmobiledevicesisgenerallythebottleneckofmultimediamobileapplications(e.g.,3Dgamesandvirtualreality).PreviousattemptstotackletheissuemainlymigrateGPUcomputationtoserversresidinginremotecloudcenters.However,thecostlynetworkdelayisespeciallyundesirableforhighly-interactivemultimediaapplicationssinceafastresponsetimeiscriticalforuserexperience.Inthispaper,weproposeGBooster,asystemthatacceleratesmultimediamobileapplicationsbytransparentlyoffloadingGPUtasksontoneighboringmultimediadevicessuchasSmartTVsandGamingConsoles.Specifically,GBoosterinterceptsandredirectssystemgraphicscallsbyutilizingtheDynamicLinkerHookingtechnique,whichrequiresnomodificationoftheapplicationsandthemobilesystems.Inaddition,amajorconcernforoffloadingisthehighenergyconsumptionincurredbynetworktransmissions.Toaddressthisconcern,GBoosterisdesignedtointelligentlyswitchbetweenthelow-powerBluetoothandthehigh-throughputWiFibasedonthetrafficdemand.WeimplementGBoosterontheAndroidsystemandevaluateitsperformance.Theresultsdemonstratethatitcanboostapplications'frameratesbyupto85%.Intermsofpowerconsumption,GBoostercanpreserveupto70%energycomparedwithlocalexecution.
Scalingk-NearestNeighborsQueries(Therightway) AtoshumSamuelCahsai(UniversityOfGlasgow),NikosNtarmos(UniversityOfGlasgow),ChristosAnagnostopoulos(UniversityofGlasgow),PeterTriantafillou(UniversityOfGlasgow)
Recentlyparallel/distributedprocessingapproacheshavebeenproposedforprocessingk-NearestNeighbours(kNN)queriesoververylarge(multidimensional)datasetsaimingtoensurescalability.However,thisistypicallyachievedattheexpenseofefficiency.Withthispaperweofferanovelapproachthatalleviatestheperformanceproblemsassociatedwithstateoftheartmethods.Theessenceofourapproach,whichdifferentiatesitfromrelatedresearch,restson(i)adoptingacoordinator-baseddistributedprocessingalgorithm,insteadofthoseemployedoverdata-parallelexecutionengines(suchasHadoop/MapReduceorSpark),and(ii)onawaytoorganizedata,tostructurecomputation,andtoindexthestoreddatasetsthatensuresthatonlyaverysmallnumberofdataitemsareretrievedfrom
![Page 83: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/83.jpg)
83
theunderlyingdatastore,communicatedoverthenetwork,andprocessedbythecoordinatorforeverykNNquery.Ourapproachalsopaysspecialattentiontoensuringscalabilityinadditiontolowqueryprocessingtimes.Overall,kNNqueriescanbeprocessedinjusttensofmilliseconds(asopposedtothe(tensof)secondsrequiredbystateoftheart.Wehaveimplementedourapproach,usingaNoSQLDB(HBase)asthedatastore,andwecompareitagainstthestate-of-the-art:theHadoop-basedSpatialHadoop(SHadoop)andtheSpark-basedSimbamethods.Weemploydifferentdatasetsofvarioussizes,showcasingthecontributedperformanceadvantages.Ourapproachoutperformsthestateoftheart,by2-3ordersofmagnitude,andconsistentlyfordatasetsizesrangingfromhundredsofmillionstohundredsofbillionsofdatapoints.Wealsoshowthatthekeyconstituentperformanceoverheadsincurredduringqueryprocessing(suchasthenumberofdataitemsretrievedfromthedatastore,therequirednetworkbandwidth,andtheprocessingtimeatthecoordinator)scaleverywell,ensuringtheoverallscalabilityoftheapproach.
ParallelizingBigDeBruijnGraphConstructiononHeterogeneousProcessors ShuangQiu(TheHongKongUniversityofScienceandTechnology),QiongLuo(TheHongKongUniversityofScienceandTechnology)
DeBruijngraphconstructionisthefirststepindenovoassemblerstoconnectinputreadsintoacompletesequencewithoutareferencegenome.Thisstepisbothtimeandmemoryspaceconsuming.Toaddressthisproblem,wedevelopParaHash,asystemthatpartitionstheinputdatainacompactformat,parallelizesthecomputationonboththeCPUsandtheGPUsinasinglecomputer,andperformshash-basedDeBruijngraphconstruction.Thisway,ParaHashutilizesallavailableprocessorstoassemblebiggenomesthatcannotfitintomemory.Furthermore,weanalyzethecharacteristicsofgenomedatatosetthehashtablesize,designconcurrenthashingalgorithmstohandletheinherentmultiplicity,andpipelinethedatatransferandthecomputationforfurtherefficiency.Ourexperimentsonreal-worldgenomedatasetsshowthattheworkloadwasbalancedacrossheterogeneousprocessors,andthatParaHashwasabletoconstructbillion-nodegraphsonasinglemachinewithanoverallperformanceupto20timesfasterthanthestate-of-the-artshared-memoryassemblers.
Private,yetPractical,MultipartyDeepLearning XinyangZhang(LehighUniversity),ShoulingJi(ZhejiangUniversity),HuiWang(StevensInstituteofTechnology),TingWang(LehighUniversity)
Inthispaper,weconsidertheproblemofmultipartydeeplearning(MDL),whereinautonomousdataownersjointlytrainaccuratedeepneuralnetworkmodelswithoutsharingtheirprivatedata.Wedesign,implement,andevaluate$\propto$MDL,anewMDLparadigmbuiltuponthreeprimitives:asynchronousoptimization,lightweighthomomorphicencryption,andthresholdsecretsharing.Comparedwithpriorwork,$\propto$MDLdepartsinsignificantways:a)besidesprovidingexplicitprivacyguarantee,itretainsdesirablemodelutility,whichisparamountforaccuracy-criticaldomains;b)itprovidesanintuitivehandlefortheoperatortogracefullybalancemodelutilityandtrainingefficiency;c)moreover,itsupportsdelicatecontrolovercommunicationandcomputationalcostsbyofferingtwovariants,operatingunderlooseandtightcoordinationrespectively,thusoptimizableforgivensystemsettings(e.g.,limitedversussufficientnetworkbandwidth).Throughextensiveempiricalevaluationusingbenchmarkdatasetsanddeeplearningarchitectures,wedemonstratetheefficacyof$\propto$MDL.
Application7:DistributedMiddlewareSystemsFastandFlexibleNetworkingforMessage-orientedMiddleware LarsKroll(KTHRoyalInstituteofTechnology),AlexandruA.Ormenisan(KTHRoyalInstituteofTechnology),JimDowling(KTHRoyalInstituteofTechnology)
Distributedapplicationsdeployedinmulti-datacenterenvironmentsneedtodealwithnetworkconnectionsofvaryingquality,includinghighbandwidthandlowlatencywithinadatacenterand,morerecently,highbandwidthandhighlatencybetweendatacentres.Inprinciple,foragivennetworkconnection,eachmessageshouldbesentoverthebestavailablenetworkprotocol,butexistingmiddlewaresdonotprovidethisfunctionality.Inthispaper,wepresentKompicsMessaging,amessagingmiddlewarethatallowsforfine-grainedcontrolofthenetworkprotocolusedonaper-messagebasis.Ratherthanalwaysrequiringapplicationdeveloperstospecifytheappropriateprotocolforeachmessage,wealsoprovideanonlinereinforcementlearnerthatoptimisestheselectionofthenetworkprotocolforthecurrentnetworkenvironment.Inexperiments,weshowhowconnectionproperties,suchasthevaryinground-triptime,influencetheperformanceoftheapplicationandweshowhowthroughputandlatencycanbeimprovedbypickingtherightprotocolattherighttime.
TailCut:PowerReductionunderQualityandLatencyConstraintsinDistributedSearchSystems Chih-HsunChou(UniversityofCalifornia,Riverside),LaxmiBhuyan(UniversityofCalifornia,Riverside),ShaoleiRen(UniversityofCalifornia,Riverside)
Websearchconstitutesanimportantclassofdataintensiveonlineservicesindatacenters.Optimizingsearchsystemsforenergyefficiency,timelyresponseandhighsearchquality(i.e.,howrelevantthereturnedresultsaretoasearchquery),however,isverychallenging,asasearchsysteminvolvesadistributedarchitecturewithhundredsofthousandsofindexservingnodes(ISNs)thatreturnsearchingresultstoanaggregatorthroughmultipleinterdependentretrievalstagesinapartition-aggregatefashion.Inthispaper,wediscoverthroughexperimentstwoimportantcharacteristicsthatcanaffectthesystemperformance:(1)responsetimeandenergyconsumptionaregreatlyimpactedbyasmallfractionofquerieswithlongprocessingtimes;(2)thequalitycontributionoftheISNisindependentofthequeryprocessingtime.Basedonourobservation,weproposeTailCut,whichjudiciouslydiscardslongqueryexecutionsandenablesISN-aggregatorcoordinationtominimizeenergyconsumptionsubjecttolatencyandqualityconstraints.OurexperimentalresultsshowthatTailCutcanachieveupto39%powersaving,whilesatisfyingthetaillatencyandqualityconstraint.
StoArranger:EnablingEfficientUsageofCloudStorageServicesonMobileDevices YongshuBai(SUNYBinghamton),YifanZhang(SUNYBinghamton)
Cloudstorageusagesarebecomingincreasinglypopularonmobiledevices.Throughanextensivemotivationstudy,wefindthatcloudstorageaccessesfrommobileappssufferfromseveralnotableproblemsthatundermineusageexperiences.Therootcauseisthatthewayofcloudstorageprovidersdeployingtheirservicesontomobiledevicesreliesonappdevelopersforthecorrectandappropriateimplementationsandlackstheabilityofmonitoringandservicingclient-sidecloudstorageaccesses.WeproposeStoArranger,apracticalsystemframeworkthatsolvestheproblemsbycoordinating,rearranging,andtransformingcloudstoragecommunicationsonmobiledevices.Wehaveprototypedtheproposedsystemusingtwodifferentimplementationapproaches.Wediscussourexperiencesoftheimplementationsinthepaper.Thereal-appevaluationexperimentsshowthatStoArrangercansignificantlyimprovemobilecloudstorageaccessefficiencywithlittleoverheads.
CharacterizingPerformanceandEnergy-efficiencyofTheRAMCloudStorageSystem
![Page 84: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/84.jpg)
84
YacineTaleb(Inria),ShadiIbrahim(Inria),GabrielAntoniu(Inria),ToniCortes(BarcelonaSupercomputingCenter)
Mostlargepopularwebapplications,likeFacebookandTwitter,havebeenrelyingonlargeamountsofin-memorystoragetocachedataandofferalowresponsetime.Asthemainmemorycapacityofclustersandcloudsincreases,itbecomespossibletokeepmostofthedatainthemainmemory.Thismotivatestheintroductionofin-memorystoragesystems.Whilepriorworkhasfocusedonhowtoexploitthelow-latencyofin-memoryaccessatscale,thereisverylittlevisibilityintotheenergy-efficiencyofin-memorystoragesystems.Eventhoughitisknownthatmainmemoryisafundamentalenergybottleneckincomputingsystems(i.e.,DRAMconsumesupto40%ofaserver’spower).Inthispaper,bythemeansofexperimentalevaluation,wehavestudiedtheperformanceandenergy-efficiencyofRAMCloud—awell-knownin-memorystoragesystem.WerevealthatalthoughRAMCloudisscalableforread-onlyapplications,itexhibitsnon-proportionalpowerconsumption.WealsofindthatthecurrentreplicationschemeimplementedinRAMCloudlimitstheperformanceandresultsinhighenergyconsumption.Surprisingly,weshowthatreplicationcanalsoplayanegativeroleincrash-recovery.
ProactivelySecureCloud-EnabledStorageKarimEldefrawy(HughesResearchLab),TylerKaczmarek(UniversityofCalifornia,Irvine),SkyFaber(UniversityofCalifornia,Irvine)
Attackingcloud-enabledstorageisbecomingincreasinglylucrativeasmorepersonalandenterprisedatamovestothecloud.Traditionalsecuritymechanismstemporarilylimitsuchattacks,butoveralongperiodoftimeattackerswilleventuallyfindvulnerabilities;thiscanleadtocompromisinglargeamountsofvaluabledataandleadtolarge-scaleprivacybreaches.Thispaperaddressesthisproblembyincorporatingproactivesecurityguaranteesintocloud-enabledstorage.Proactivesecuritydealswithanadversary’sabilitytoeventuallycompromiseallinvolvedserversinadistributedstorageorcomputationsystem.Whilethereareseveralproactivelysecuresecretsharingprotocolsthatcanbeusedtoimproveconfidentialityofdatastoredinthecloud,theirhighoverheadhastraditionallylimitedthemtolessthantenpartiesandtoonly100sofbytestypicalforcryptographickeys.Realizingproactivelysecurecloudstorageforlargerdata(e.g,MBs)requirescarefuldesignandcalibrationofsystemparameters,andfacesseveralchallenges.Inthispaperwedesign,implementandassessperformanceofthefirstsystemforProactivelySecureCloud-EnabledStorage(PiSCES)ofdatalargerthancryptographickeys.Basedonourpracticalperformanceresultsweadvocatethatthehighlevelofresilienceandlong-termsecurityandconfidentialityguaranteesenabledbyproactivesecurityshouldbeconsideredinfuturedistributedandcloud-basedstorageandcomputingservices.
BEES:Bandwidth-andEnergy-EfficientImageSharingforReal-timeSituationAwareness PengfeiZuo(HuazhongUniversityofScienceandTechnology),YuHua(HuazhongUniversityofScienceandTechnology),XueLiu(McGillUniverisity),DanFeng(HuazhongUniversityofScienceandTechnology),WenXia(HuazhongUniversityofScienceandTechnology),ShundeCao(HuazhongUniversityofScienceandTechnology),JieWu(HuazhongUniversityofScienceandTechnology),YuanyuanSun(HuazhongUniversityofScienceandTechnology),YunchengGuo(HuazhongUniversityofScienceandTechnology)
Inordertosavehumanlivesandreduceinjuryandpropertyloss,SituationAwareness(SA)informationisessentialandimportantforrescueworkerstoperformtheeffectiveandtimelydisasterrelief.Theinformationisgenerallyderivedfromthesharedimagesviawidelyusedsmartphones.However,conventionalsmartphone-basedimagesharingschemesfailtoefficientlymeettheneedsofSAapplicationsduetotwomainreasons,i.e.,real-timetransmissionrequirementandapplication-levelimageredundancy,whichisexacerbatedbylimitedbandwidthandenergyavailability.Inordertoprovideefficientimagesharingindisasters,weproposeabandwidth-andenergy-efficientimagesharingsystem,calledBEES.ThesalientfeaturebehindBEESistoproposetheconceptofApproximateImageSharing(AIS),whichexploresandexploitsapproximatefeatureextraction,redundancydetection,andimageuploadingtotradetheslightlylowqualityofcomputationresultsincontent-basedredundancyeliminationforhigherbandwidthandenergyefficiency.Nevertheless,theboundariesofthetradeoffsbetweenthequalityofcomputationresultsandefficiencyaregenerallysubjectiveandqualitative.Wehenceproposetheenergy-awareadaptiveschemesinAIStoleveragethephysicalenergyavailabilitytoobjectivelyandquantitativelydeterminethetradeoffsbetweenthequalityofcomputationresultsandefficiency.Moreover,unlikeexistingworkonlyforcross-batchsimilarimages,BEESfurthereliminatesin-batchonesviaasimilarity-awaresubmodularmaximizationmodel.WehaveimplementedtheBEESprototypewhichisevaluatedviathreereal-worldimagedatasets.ExtensiveexperimentalresultsdemonstratetheefficacyandefficiencyofBEES.
Application8:DistributedSystemsandOptimizationsTransparentFault-ToleranceusingIntra-MachineFull-Software-StackReplication GiulianoLosa(VirginiaTech),AntonioBarbalace(VirginiaTech),YuzhongWen(VirginiaTech),MarinaSadini(VirginiaTech),Ho-RenChuang(VirginiaTech),BinoyRavindran(VirginiaTech)
Asthenumberofprocessorsandthesizeofthememoryofcomputingsystemskeepincreasing,thelikelihoodofCPUcorefailures,memoryerrors,andbusfailuresincreasesandcanthreatensystemavailability.SoftwarecomponentscanbehardenedagainstsuchfailuresbyrunningseveralreplicasofacomponentonhardwarereplicasthatfailindependentlyandthatarecoordinatedbyaState-MachineReplicationprotocol.Onecommonsolutionistoreplicatethephysicalmachinetoprovideredundancy,andtorewritethesoftwaretoaddresscoordination.However,aCPUcorefailure,amemoryerror,orabuserrorisunlikelytoalwayscrashanentiremachine.Thus,fullmachinereplicationmaysometimesbeanoverkill,increasingresourcecosts.Inthispaper,weintroducefullsoftwarestackreplicationwithinasinglecommoditymachine.Ourapproachrunsreplicasonfault-independenthardwarepartitions(e.g.,NUMAnodes),whereineachpartitionissoftware-isolatedfromtheothersandhasitsownCPUcores,memory,andfullsoftwarestack.Ahardwarefailureinonepartitioncanberecoveredbyanotherpartitiontakingoveritsfunctionality.WehaverealizedthisvisionbyimplementingFT-Linux,aLinux-basedoperatingsystemthattransparentlyreplicatesrace-free,multithreadedPOSIXapplicationsondifferenthardwarepartitionsofasinglemachine.OurevaluationsofFT-LinuxonseveralpopularLinuxapplicationsshowaworstcaseslowdown(duetoreplication)by~20%.
Apreventiveauto-parallelizationapproachforelasticstreamprocessing RolandKottoKombi(UniversityClaudeBernard),NicolasLumineau(UniversitédeLyon),PhilippeLamarre(INSALyon)
Nowadays,moreandmoresources(connecteddevices,socialnetworks,etc.)emitreal-timedatawithfluctuatingratesovertime.Existingdistributedstreamprocessingengines(SPE)havetoresolveadifficultproblem:deliverresultssatisfyingend-usersintermsofqualityandlatencywithoutover-consumingresources.Thispaperfocusesonparallelizationofoperatorstoadapttheirthroughputtotheirinputrate.Wesuggestanapproachwhichpreventsoperatorcongestioninordertolimitdegradationofresultsquality.Thisapproachreliesonanautomaticanddynamicadaptationofresourceconsumptionforeachcontinuousquery.Thissolutiontakesadvantageofi)ametricestimatingtheactivitylevelofoperatorsinthenearfutureii)theAUTOSCALEapproachwhichevaluatestheneedtomodifyparallelismdegreesatlocalandglobalscopeiii)anintegrationintotheApacheStormsolution.WeshowperformancetestscomparingourapproachtothenativesolutionofthisSPE.
![Page 85: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/85.jpg)
85
DependableCloudResourceswithGuardian BaraAbusalah(PurdueUniversity),DerekSchatzlein(PurdueUniversity),JulianJamesStephen(PurdueUniversity),MasoudSaeidaArdekani(PurdueUniversity),PatrickEugster(PurdueUniversity)
Despiteadvancesinmakingdatacentersdependable,failuresstillhappen.Thisisparticularlyonerousforlong-running“bigdata”applications,wherepartialfailurescanleadtosignificantlossesandlengthyrecomputations.BigdataprocessingframeworkslikeHadoopMapReduceincludefaulttolerance(FT)mechanisms,butthesearecommonlytargetedatspecificsystem/failuremodels,andareoftenredundantbetweenframeworks.Thispaperproposestheparadigmofdependableresources:bigdataprocessingframeworksaretypicallybuiltontopofresourcemanagementsystems(RMSs),andproposingFTsupportatthelevelofsuchanRMSyieldsgenericFTmechanisms,whichcanbeprovidedwithlowoverheadbyleveragingconstraintsonresources.WedemonstrateourconceptsthroughGuardian,arobustRMSbasedonYARN.GuardianallowsframeworkstoruntheirapplicationswithindividuallyconfigurableFTgranularityanddegree,withonlyminorchangestotheirimplementation.WedemonstratethebenefitsofourapproachbyevaluatingHadoop,Tez,SparkandPigonGuardianinAmazon-EC2,improvingcompletiontimebyaround68%inthepresenceoffailures,whilemaintainingaround6%overhead.
ACommunication-awareContainerRe-distributionApproachforHighPerformanceVNFs YuchaoZhang(TsinghuaUniversity),YusenLi(NankaiUniversity),KeXu(TsinghuaUniversity),DanWang(HongKongPolytechnicUniversity),MinghuiLi(Baidu),XuanCao(Baidu),QingqingLiang(Baidu)
Containershavebeenusedinmanyapplicationsforisolationpurposesduetothelightweight,scalableandhighlyportableproperties.However,toapplycontainersinvirtualnetworkfunctions(VNFs)facesabigchallengebecausehigh-performanceVNFsoftengeneratefrequentcommunicationworkloadsamongcontainerswhilethecontainercommunicationsaregenerallynotefficient.Comparedwithhardwaremodificationsolutions,properlydistributingcontainersamonghostsisanefficientandlow-costwaytoreducecommunicationoverhead.However,weobservethatthisapproachyieldsatrade-offbetweenthecommunicationoverheadandtheoverallthroughputofthecluster.Inthispaper,wefocusonthecommunication-awarecontainerredistributionproblemtooptimizethecommunicationoverheadandtheoverallthroughputjointlyforVNFclusters.WeproposeasolutioncalledFreeContainerwhichutilizesanoveltwo-stagealgorithmtore-distributecontainersamonghosts.WeimplementFreeContainerinBaiduclusterswith6000serversand35servicesdeployed.Extensiveexperimentsonrealnetworksareconductedtoevaluatetheperformanceoftheproposedapproach.TheresultsshowthatFreeContainercanincreasetheoverallthroughputupto90%withsignificantreductiononcommunicationoverhead.
MinimizingCostinIaaSCloudsviaScheduledInstanceReservation QiushiWang(NanyangTechnologicalUniversity),MingMingTan(NanyangTechnologicalUniversity),XueyanTang(NanyangTechnologicalUniversity),WentongCai(NanyangTechnologicalUniversity)
Regulardiurnalpatternsareoftenseenintheworkloadsofcloud-basedonlineapplications.Thiskindofnon-stationaryworkloadschangestheprocessingdemandsovertime.Torunapplicationserviceswithminimumcosts,thenumberofcloudinstancescanbedynamicallyadjustedaccordingtotheworkloadvariations.Recently,anewtypeofscheduledinstanceshasemergedintheInfrastructure-as-a-Servicemarkettofacilitatesuchconfigurations.Scheduledinstancescanbereservedbasedonarecurringscheduleandtheyofferpricediscounts.Meanwhile,cloudvendorsrequireminimumscheduleddurationstoavoidtheoverheadoffrequentlylaunchingandterminatingcloudinstances.Coupledwithtraditionalon-demandandreservedinstances,itbecomesmorecomplicatedforuserstofindtheoptimalcombinationofthesethreepricingoptionstominimizetheirmonetarycosts.Forthenewscheduledinstances,notonlythenumberofinstancesbutalsotheirstartandstoptimeshavetobedecided.Inthispaper,wedevelopafastandeffectivestrategytosolvethisproblem.Basedonthehourlyworkloaddistributions,wefirstcomputetheoptimalnumberofinstancestoacquireforeachpricingoption.Then,wedesignaschedulingalgorithmtoarrangethescheduledinstancesincompliancewiththerestrictionoftheirscheduleddurations.UsingtheworkloadsoftheLOLonlinegameandtheWikipediaMobileserviceastwocasestudies,theefficacyofourstrategyisdemonstrated.
EfficientDistributedCoordinationatWAN-scale AilidaniAilijiang(SUNYBuffalo),AlekseyCharapko(SUNYBuffalo),MuratDemirbas(SUNYBuffalo),BekirOguzTurkkan(SUNYBuffalo),TevfikKosar(SUNYBuffalo)
Traditionalcoordinationservicesfordistributedapplicationsdonotscalewelloverwide-areanetworks(WAN):centralizedcoordinationfailstoscalewithrespecttotheincreasingdistancesintheWAN,anddistributedcoordinationfailstoscalewithrespecttothenumberofnodesinvolved.WearguethatitispossibletoachievescalabilityoverWANusingahierarchicalcoordinationarchitectureandasmarttokenmigrationmechanism,andlaydownthefoundationofanoveldesignforaflexible-consistentcoordinationframework,calledWanKeeper.WeimplementedWanKeeperbasedontheZooKeeperAPIanddeployeditoverWANasaproofofconcept.OurexperimentalresultsbasedontheYahoo!CloudServingBenchmark(YCSB),ApacheBookKeeperreplicatedlogservice,andtheSharedCloud-backedFileSystem(SCFS)showthatWanKeeperprovidesmultiplefoldsimprovementinwrite/updateperformanceinWANcomparedtoZooKeeper,whilekeepingthesamereadperformance.
Application9:DistributedSystemsandApplicationsSpecifyingaDistributedSnapshotAlgorithmasaMeta-programandModelCheckingitatMeta-level HaThiThuDoan(JapanAdvancedInstituteofScienceandTechnology),FrancoisBonnet(OsakaUniversity),KazuhiroOgata(JapanAdvancedInstituteofScienceandTechnology)
ThepaperproposesanewapproachtomodelcheckingChandy-LamportDistributedSnapshotAlgorithm(CLDSA).TheessentialoftheapproachisthatCLDSAisspecifiedasameta-programinMaudesuchthatthemeta-programtakesaspecificationofanunderlyingdistributedsystem(UDS)andgeneratesthespecificationoftheUDSonwhichCLDSAissuperimposed(UDS-CLDSA).TomodelcheckthataUDS-CLDSAenjoysadesiredproperty,itsufficesthathumanusersspecifytheUDSfortheproposedapproach,whilehumanusersneedtospecifytheUDS-CLDSAfortheexistingapproachforeachUDS.Sincetheproposedapproachconductsmodelcheckingatmeta-level,itproducesacounterexampleifaUDS-CLDSAdoesnotenjoytheproperty,whiletheexistingapproachdoesnot.OurmethodspecifyingCLDSAasameta-programcanbeappliedtoformalspecificationoftheclassofdistributedalgorithmsthataresuperimposedonUDSs.
Self-EvolvingSubscriptionsforContent-BasedPublish/SubscribeSystems
![Page 86: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/86.jpg)
86
CesarCañas(McGillUniversity),KaiwenZhang(TechnicalUniversityofMunich),BettinaKemme(McGillUniversity),JörgKienzle(McGillUniversity),Hans-ArnoJacobsen(TechnicalUniversityofMunich)
Traditionalpub/subsystemscannotadequatelyhandleworkloadsofapplicationswithdynamic,short-livedsubscriptionssuchaslocation-basedsocialnetworks,predictivestocktrading,andonlinegames.Subscribersmustcontinuouslyinteractwiththepub/subsystemtoremoveandinsertsubscriptions,therebyinefficientlyconsumingnetworkandcomputingresources,andsacrificingconsistency.Intheaforementionedapplications,werecognizethatthechangesinthesubscriptionscanfollowapredictablepatternoversomevariable(e.g.,time).Inthispaper,wepresentanewtypeofsubscription,calledevolvingsubscription,whichencapsulatesthesepatternsandallowthepub/subsystemtoautonomouslyadapttothedynamicinterestsofthesubscriberswithoutincurringanexpensivere-subscriptionoverhead.Weproposeageneralmodelforexpressingevolvingsubscriptionsandaframeworkforsupportingtheminapub/subsystem.Tothisend,weproposethreedifferentdesignstosupportevolvingsubscriptions,whichareevaluatedandcomparedtothetraditionalresubscriptionapproachinthecontextoftwousecases:onlinegamesandhigh-frequencytrading.Ourevaluationshowsthatoursolutionscanreducesubscriptiontrafficby96.8%andimprovedeliveryaccuracywhencomparedtothebaselineresubscriptionmechanism.
ScalableRoutingforTopic-basedPublish/SubscribeSystemsunderFluctuations VolkerTurau(HamburgUniversityofTechnology),GerrySiegemund(HamburgUniversityofTechnology)
Theloosecouplingandtheinherentscalabilitymakepublish/subscribesystemsanidealcandidateforevent-drivenservicesforwirelessnetworksusinglowpowerprotocolssuchasIEEE802.15.4.Thisworkintroducesadistributedalgorithmtobuildandmaintainaroutingstructureforsuchnetworks.Thealgorithmdynamicallymaintainsamulticasttreeforeachnode.Whilepreviousworkfocusedonminimizingthesetreesweaimtokeeptheefforttomaintainthemincaseoffluctuationsofsubscriberslow.Themulticasttreesareimplicitlydefinedbyanovelstructurecalledaugmentedvirtualring.Themaincontributionisadistributedalgorithmtobuildandmaintainthisaugmentedvirtualring.Maintenanceoperationsaftersub-andunsubscriptionsrequiremessageexchangeinalimitedregiononly.Wecomparetheaveragelengthsoftheconstructedforwardingpathswithanalmostidealapproach.AsaresultofindependentinterestwepresentadistributedalgorithmusingmessagesofsizeO(logn)forconstructingvirtualringsofgraphsthatareonaverageshorterthanringsbasedondepthfirstsearch.
OPPay:DesignandImplementationofAPaymentSystemforOpportunisticDataServices FengruiShi(ImperialCollegeLondon),ZhijinQin(ImperialCollegeLondon),JulieMcCann(ImperialCollegeLondon)
Thelargenumberofpersonalwirelessdevicesintheurbanareascouldbeusedtoprovidevariousopportunisticdataservices,suchasWiFisharing,content-basedfilesharingandopportunisticnetworking.Inordertofacilitatetheseservices,itisessentialtoincentivisethedeviceownerstobecomeserviceproviders.However,previousresearchfailedtodeliveranypracticalpaymentsystemsforopportunisticdataservices.Inspiredbysmartcontractsfunctionalitiesofbitcoin,thispaperproposesapaymentsystemnamedOPPayforopportunisticdataservices,whichimplementsamicropaymentcommunicationprotocolformobiledevicestoperformdatatransactionsandmakepaymentsusingbitcoin.Thesystemisdesignedtomakeincrementalpaymentsandthusresilienttointerruptedcommunicationscausedbyhumanmobilityinthemobilenetwork.Byimplementingandevaluatingthesystemforthreedifferentapplications,weshowthatthesystemisabletoworkinheterogeneoushardwareandsoftwareenvironmentsandcanachievefasttransactionsconfirmationwithsmallfeeoverheadandlowfaultypaymentvalue.
OptimalResourceAllocationforMulti-userVideoStreamingovermmWaveNetworksZhifengHe(AuburnUniversity),ShiwenMao(AuburnUniversity)
Weinvestigatetheresourceallocationproblem,includingtimeslotallocation,channelallocation,andpoweradaptation,inamillimeterWave(mmWave)networkwithmultipletransmissionlinks,multiplechannels,andaPicoNetCoordinator(PNC).Eachlinkhasavideosessiontotransmitfromthetransmittertothereceiver.Theobjectiveistominimizethenumberoftimeslotstofinishthevideosessionsofalllinksbyjointlyoptimizingchannelallocationandtimeslotallocationforlinks,whileconsideringthepossibleinterferencebetweendifferentlinksonthesamechannel.Theoptimalsolutionfortheformulatedproblemiscomputationallyprohibitivetoobtainduetotheexponentialcomplexity.Wedevelopedacolumngenerationbasedmethodtoreformulatetheoriginalproblemintoamainproblemalongwithaseriesofsub-problems,withgreatlyreducedcomplexity.Weprovethattheoptimalsolutionforthereformulatedproblemconvergestotheoptimalsolutionoftheoriginalproblem,andwederivedalowerboundfortheperformanceofthereformulatedproblemateachiteration,whichwillfinallyconvergetotheglobaloptimalsolution.Theproposedschemeisvalidatedwithsimulationswithitssuperiorperformanceoverexistingworkisobserved.
AMulti-AgentParallelApporachtoAnalyzingLargeClimateDataSets JasonWoodring(UniversityofWashingtonBothell),MatthewSell(UniversityofWashingtonBothell),MunehiroFukuda(UniversityofWashingtonBothell),HazelineAsuncion(UniversityofWashingtonBothell),EricSalathe(UniversityofWashingtonBothell)
Despitevariouscloudtechnologiesthathaveparallelizedandscaledupbigdataanalysis,theytargetdatamostlyintextswhichareeasytopartitionandthuseasytomapoveraclustersystem.Therefore,theirparallelizationdonotnecessarilycoverscientificstructureddatasuchasNetCDForneedadditional,user-providedtoolstoconverttheoriginaldataintospecificformats.Tofacilitateuser-intuitiveparallelizationofsuchscientificdataanalysis,thispaperpresentsanagent-basedapproachthatinstantiatesdistributedarraysoveraclustersystem,maintainsstructuredscientificdatainthesearrays,deploysmanymobileagentsoverthearraystoperformcomputationalactionsondata,andcollectsnecessaryresults.Todemonstratethepracticabilityofouragent-basedapproach,wefocusedonclimatechangeresearchandimplementedaweb-interfacedclimateanalysis,usingtheMASS(multi-agentspatialsimulation)library.Inthispaper,weshowpracticaladvantagesof,performanceimprovementsby,andchallengesforouragent-basedapproachinstructureddataanalysis.
Application10:DistributedSystemsandServicesEnergyProportionalServers:WhereAreWein2016? CongfengJiang(HangzhouDianziUniversity),YumeiWang(HangzhouDianziUniversity),DongyangOu(HangzhouDianziUniversity),BingLuo(WayneStateUniversity),WeisongShi(WayneStateUniversity)
Thehugeenergyconsumptionindatacentersproducesnotonlyhighelectricitybillbutalsotremendouscarbonfootprints.Althoughtoday’sserversanddatacentersofleadinginternetcompaniesaremoreenergyefficientthaneverbefore,thefluctuationsinexternalworkloadandinternalresourceutilizationcallsforenergyproportionalcomputing.Insightintoserverenergyproportionalitycanhelpimproveworkloadplacementwhilealsoreducingenergyconsumption.Inthispaper,weinvestigateall477validpublishedresultsofSPECpower_ssjbenchmarkfrom2007to2016Q3andreorganizethembyhardwareavailabilityyearformoreaccurateanalysisonproductionservers.Throughcomprehensiveanalysiswefindthat:(1)Thespeciousstagnationofenergyproportionalityinrecentyearsis
![Page 87: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/87.jpg)
87
mainlycausedbytheadoptionofprocessorsofspecificmicroarchitectureandisnottheindicativetrendofenergyproportionalityimprovement.(2)Microarchitectureevolutionhasmoreinfluenceonenergyefficiencyimprovementthanenergyproportionality.(3)Today’sservers’peakenergyefficienciesareshiftingfrom100%resourceutilizationto80%or70%utilizationandserverenergyproportionalityimproveswithsuchshifting.Wethenconductextensiveexperimentson4rackserverstoinvestigatetheenergyefficiencyvariationsunderdifferenthardwareconfigurations,includingmemorypercoreinstallationandprocessorfrequencyscaling.Ourexperimentsshowthathardwareconfigurationhassignificantimpactonserver’senergyefficiency.Ourfindingspresentedinthispaperprovideusefulinsightsandguidancetosystemdesigners,aswellasdatacenteroperatorsforenergyproportionalityawareworkloadplacementandenergysavings.
AreHTTP/2ServersReadyYet? MuhuiJiang(TheHongKongPolytechnicUniversity),XiapuLuo(TheHongKongPolytechnicUniversity),TungngaiMiu(NexusguardLimited),ShengtuoHu(TheHongKongPolytechnicUniversity),WeixiongRao(TongjiUniversity)
SupersedingHTTP/1.1,thedominatingwebprotocol,HTTP/2promisestomakewebapplicationsfasterandsaferbyintroducingmanynewfeatures,suchasmultiplexing,headercompression,requestpriority,serverpush,etc.AlthoughafewrecentstudiesexaminedtheadoptionofHTTP/2andevaluateditsimpacts,littleisknownaboutwhetherthepopularHTTP/2servershavecorrectlyrealizedthenewfeaturesandhowthedeployedserversusethesefeatures.Tofillinthegap,inthispaper,weconductthefirstsystematicinvestigationbyinspectingsixpopularimplementationsofHTTP/2servers(i.e.,Nginx,Apache,H2O,Lightspeed,nghttpdandTengine)andmeasuringthetop1millionAlexawebsites.Inparticular,weproposenewmethodsanddevelopatoolnamedH2Scopetoassessthenewfeaturesinthoseservers.Theresultsofthelarge-scalemeasurementonHTTP/2websitesrevealnewobservationsandinsights.ThisstudyshedslightonthecurrentstatusandthefutureresearchofHTTP/2.
DataIntegrityforCollaborativeApplicationsoverHostedServices ErtemEsiner(NanyangTechnologicalUniversity),AnwitamanDatta(NanyangTechnologicalUniversity)
Inthisworkwefocusonintegrityandconsistencyofdataaccessedandmanipulatedbymultiplecollaboratingusers,andstoredinan(untrusted)hostedservice.Thisisaproblem,aspectsofwhichhavebeenstudiedinisolationinhithertodistinctcommunities.Consistencyisoneofthecardinalproblemsofdistributedcomputing.Integrityofhosteddatahasbeenstudiedoverthelastdecade,andnumeroustechniquesforproofofdatapossessionand/orretrievabilityhavebeenexplored.Thelatterlineofworkhoweverhaveoftenassumedstaticdata,andtechniquestohandledynamicorversioneddatahaveonlyveryrecentlybeenproposed.Yet,eventheexistingsolutionsthathandlemutablecontentdosoundertheassumptionthatonlyasingledataowner(usingasingleclient)manipulateandverifysaiddata.Thisisaseriouslimitationintermsofthevarietyofapplicationsthatcanbenefitfromsuchmechanismsforproofofdatapossession.Thenovelty,andprimarycontributionofthisworkisinfillingthisgap.Specifically,weextendtheexistingideasofproofofpossessionofdynamicdata,inordertosupportmultipleuserswhomaycollaborateinrealtimeorasynchronously.Incontrast(andaddition)tothechallengeofanuntrustedstorageserverthatexistingtechniquesforproofofdatapossessionneedtoovercome,wehadto,simultaneouslyaccountfordataintegrityviolationsthatmaybeincurredduetoalltheusualchallengesofmaintainingconsistencyofcollaborativedata(evenifthestorageserverwastrusted).
VirtualMachinePowerAccountingwithShapleyValue WeixiangJiang(HuazhongUniversityofScience&Technology),FangmingLiu(HuazhongUniversityofScienceandTechnology),GuomingTang(UniversityofVictoria),KuiWu(UniversityofVictoria),HaiJin(HuazhongUniversityofScience&Technology)
Theever-increasingpowerconsumptionofdatacentershaseatenupalargeportionoftheirprofit.Onepossiblesolutionistochargedatacenterusersfortheiractualpowerusage.However,itposesagreattechnicalchallengeasthepowerofVMsco-existinginaphysicalmachinecannotbemeasureddirectly.ItisthuscriticaltodevelopafairmethodtodisaggregatethepowerofaphysicalmachinetoindividualVMs.Wetackletheabovechallengebymodelingthepowerdisaggregationproblemasacooperativegameandproposenon-deterministicShapleyvaluetodiscoverthefairpowershareofVMs(inthesenseofsatisfyingfourdesiredaxiomaticprinciples),whilecompensatingthenegativeimpactofVMpowervariation.Wedemonstratethattheresultsfromexistingpowermodel-basedsolutioncandeviatefromthe“groundtruth”by25.22%-46.15%.AndcomparedwiththeexactShapleyvalue,ournon-deterministicShapleyvaluecanachievelessthan5%errorfor90%ofthetime.
AVersatilePlatformforMobileDataGatheringExperimentsinWirelessSensorNetworks JiLi(StonyBrookUniversity),CongWang(StonyBrookUniversity),YuanyuanYang(StonyBrookUniversity)
Inrecentyears,mobiledatagatheringinwirelesssensornetworkshasattractedmuchinterestsintheresearchcommunity.However,despiteextensiveefforts,manyofpreviousworkinthisarealiesonlyintheoryandevaluatesnetworkperformancewithcomputersimulations,whichleavesalargegapfromreality.Inthispaper,wepresentthedesignandimplementationofageneralpurpose,flexibleplatformformobiledatagatheringinwirelesssensornetworkstoevaluatenetworkperformanceandalgorithmsinapracticalsetting.Insteadofrelyingonhand-craftedtheoreticalmodels,ourplatformintegratesbothmobiledatacollectorandsensornodestoproviderealisticperformanceevaluations.Inaddition,theplatformadoptsamodulardesigninmobiledatacollectorandsensornodes,andequipsthemobiledatacollectorwithadvancedcomputingcapability,whichmakesitversatileforevaluatingtheperformanceofawide-rangeofapplications.Finally,asacasestudy,weimplementawildlifemonitoringsystemonourplatform.Ourexperimentalresultsdemonstratethatrealimplementationscanevaluatemanypracticalperformancefactorswhichwouldhaveagreatimpactonthesensingresultsandareverydifficulttofullycapturebytheoreticalmodelsandsimulations.Weexpectthatthisplatformcanbecomeaverypowerfulgeneraltoolformoreaccuratenetworksimulationsandfacilitateperformanceoptimizationinwirelesssensornetworks.
OnDirectionalNeighborDiscoveryinmmWaveNetworks YuWang(AuburnUniversity),ShiwenMao(AuburnUniversity),TheodoreS.Rappaport(NewYorkUniversity)
Thedirectionalneighbordiscoveryproblem,i.e.,spatialrendezvous,isafundamentalprobleminmillimeterwave(mmWave)networks.Thechallengeishowtoletthetransmitterandreceiverbeamsmeetinspaceunderdeafnesscausedbydirectionaltransmissionandreception.Inthispaper,wepresentaHunting-basedDirectionalNeighborDiscovery(HDND)scheme,whereanodecontinuouslyrotatesitsdirectionalbeamtoscanitsneighborhoodforneighbors.Througharigorousanalysis,wederivetheconditionsforensuredneighbordiscovery,aswellasaboundfortheworstcasediscoverytime.Wevalidatetheanalysiswithextensivesimulations,anddemonstratethesuperiorperformanceoftheproposedschemeovertwobenchmarkschemes.
![Page 88: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/88.jpg)
88
Vision/Blue Sky Thinking Track Paper Abstracts
Vision1:InternetofThings,SmartCitiesandCyber-PhysicalSystemsObservable-by-Design MasaruKitsuregawa(NationalInstituteofInformatics(NII)/InstituteofIndustrialScience,UniversityofTokyo)
Wepresenttheobservable-by-designprinciple.Webelieve,thatthenewgenerationofservices,products,andenvironment,managementsystemsshouldbedesignedtoadapttochanges.,Therefore,theyshouldbedesignedtobeobservable,andtheirdesign,shouldproactivelyandreactivelyadapttothechangesobserved,bothinternallyandexternally.Twoconcreteexamplesillustrate,theapplicationofobservable-by-designprinciple:(1)ship,buildingandmanagement,and(2)riverdamwaterflowmanagement.,Webelievethattheobservable-by-designprinciplecanbe,appliedinalargescale.Inthelongterm,anewgenerationofobservable-by-design,infrastructurescanbebuiltthatincorporates,thesensingandadaptingcapabilitiesintheirconstruction.
AnArchitecturalVisionforaData-CentricIoT:RethinkingThings,TrustandClouds EveM.Schooler(Intel),DavidZage(Intel),JeffSedayao(Intel),HassnaaMoustafa(Intel),AndrewBrown(Intel),MorenoAmbrosin(UniversityofPadua)
TheInternetofThings(IoT)isproducingatidalwaveofdata,muchofitoriginatingatthenetworkedgeandoriginatingfromapplicationswithrequirementsunmetbythetraditionalback-endCloudarchitecture.Toaddressthedisruptioncausedbytheoceanofdata,thispaperoffersaholisticdata-centricarchitecturalvisionforthedata-centricIoT.Itadvocatesthatwerethinkourapproachtothedesignanddefinitionofkeyelements:thatweshiftourfocusfromThingstoSmartObjects;growTrustorganically;andevolveback-endCloudstowardEdgeandFogclouds,whichleveragedata-centricnetworksandenableoptimalhandlingofupstreamdataflows.Alongtheway,wewaxpoeticaboutseveralblue-skytopics,assessthestatusoftheseelementsinthecontextofrelatedwork,andidentifyknowngapsinmeetingthisvision.
EdgeComputingandIoTBasedResearchforBuildingSafeSmartCitiesResistanttoDisasters TeruoHigashino(OsakaUniversity),HirozumiYamaguchi(OsakaUniversity),AkihitoHiromori(OsakaUniversity),AkiraUchiyama(OsakaUniversity),KeiichiYasumoto(NaraInstituteofScienceandTechnology)
Recently,severalresearchesconcerningwithsmartandconnectedcommunitieshavebeenstudied.Soonthe4G/5Gtechnologybecomespopular,andcellularbasestationswillbedenselylocatedintheurbanspace.Theymayofferintelligentservicesforautonomousdriving,urbanenvironmentimprovement,disastermitigation,elderly/disabledpeoplesupportandsoon.Suchinfrastructuremightfunctionasedgeserversfordisastersupportbase.Inthispaper,weenumerateseveralresearchissuestobedevelopedintheICDCScommunityinthenextdecadeinorderforbuildingsafesmartcitiesresistanttodisasters.Inparticular,wefocuson(A)up-to-dateurbancrowdmobilitypredictionand(B)resilientdisasterinformationgatheringmechanismsbasedontheedgecomputingparadigm.Weinvestigaterecentrelatedworksandprojects,andintroduceouron-goingresearchworkandinsightfordisastermitigation.
TheInternetofThingsandMultiagentSystems:DecentralizedIntelligenceinDistributedComputing MunindarSingh(NorthCarolinaStateUniversity),AmitChopra(LancasterUniversity)
Traditionally,distributedcomputingconcentratesoncomputationunderstoodatthelevelofinformationexchangeandsetsasidehumanandorganizationalconcernsaslargelytobehandledinanadhocmanner.Increasingly,however,distributedapplicationsinvolvemultiplelociofautonomy.Researchinmultiagentsystems(MAS)addressesautonomybydrawingonconceptsandtechniquesfromartificialintelligence.However,MASresearchgenerallylacksanadequateunderstandingofmoderndistributedcomputing.InthisBlueSkypaper,weenvisiondecentralizedmultiagentsystemsasawaytoplacedecentralizedintelligenceindistributedcomputing,specifically,bysupportingcomputationatthelevelofsocialmeanings.WemotivateourproposalsforresearchinthecontextoftheInternetofThings(IoT),whichhasbecomeamajorthrustindistributedcomputing.FromtheIoTsrepresentativeapplications,weabstractoutthemajorchallengesofrelevancetodecentralizedintelligence.TheseincludetheheterogeneityofIoTcomponents;asynchronousanddelay-tolerantcommunicationanddecoupledenactment;andmultiplestakeholderswithsubtlerequirementsforgovernance,incorporatingresourceusage,cooperation,andprivacy.TheIoTyieldshigh-impactproblemsthatrequiresolutionsthatgobeyondtraditionalwaysofthinking.Weconcludewithhighlightsofsomepossibleresearchdirectionsindecentralizedmultiagentsystems,includingprogrammingmodels;interaction-orientedsoftwareengineering;andwhatwetermenlightenedgovernance.
InternetofThings:FromSmall-toLarge-ScaleOrchestration CharlesConsel(Inria/BordeauxINP),MilanKabac(ImperialCollege)
ThedomainofInternetofThings(IoT)israpidly,expandingbeyondresearch,andbecomingamajorindustrial,marketwithsuchstakeholdersasmajormanufacturersofchips,andconnectedentities(i.e.,things),andfast-growingoperators,ofwide-areanetworks.Importantly,thisemergingdomainis,drivenbyapplicationsthatleverageanIoTinfrastructureto,provideuserswithinnovative,high-valueservices.IoTinfrastructures,rangefromsmallscale(e.g.,homesandpersonal,health)tolargescale(e.g.,citiesandtransportationsystems).Inthispaper,wearguethatthereisacontinuumbetweenorchestrating,connectedentitiesinthesmallandinthelarge.We,proposeaunifiedapproachtoapplicationdevelopment,which,coversthisspectrum.Todoso,weexaminetherequirementsfor,orchestratingconnectedentitiesandaddressthemwithdomainspecific,designconcepts.Wethenshowhowtomapthesedesign,conceptsintodedicatedprogrammingpatternsandruntime,mechanisms.Ourworkrevolvesarounddomain-specificconceptsand,notations,integratedintoatool-baseddesignmethodologyand,dedicatedtodevelopIoTapplications.Wehaveappliedour,workacrossaspectrumofinfrastructuresizes,rangingfrom,anautomatedpilotinavionics,toanassistedlivingplatform,forthehomeofseniors,toaparkingmanagementsystemin,asmartcity.
EdgeOS_H:AHomeOperatingSystemforInternetofEverything JieCao(WayneStateUniversity),LanyuXu(WayneStateUniversity),RaefAbdallah(WayneStateUniversity),WeisongShi(WayneStateUniversity)
TheproliferationofInternetofEverything(IoE)andthesuccessofrichCloudserviceshavepushedthehorizonofanewcomputingparadigm,EdgeComputing,whichcallsforprocessingthedataattheedgeofthenetwork.SmarthomeasatypicalIoEapplicationisbeingwidelyadaptedintopeopleslife.EdgeComputinghasthepotentialtoempowerthesmarthome,butitneedsmorecontributionfromthecommunitybeforeittrulybenefitsourlives.Inthispaper,wefirstpresentthe
![Page 89: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/89.jpg)
89
visionofEdgeOSH,ahomeoperatingsystemforInternetofEverything,followedbythechallengesinEdgeOSH,namelyprogramminginterface,self-management,datamanagement,security&privacy,andnaming.Withineachchallengewealsodiscussthepotentialdirectionsthatareworthfurtherinvestigation.
Vision2:FutureNetworkingandCyberinfrastructureAVisionforZero-HopNetworking(ZeN) MostafaAmmar(SchoolofComputerScience,GeorgiaTech),EllenZegura(SchoolofComputerScience,GeorgiaTech),YimengZhao(SchoolofComputerScience,GeorgiaTech)
Ithasbecomeincreasinglyimportantforcontentproviders(CPs)toreachconsumerswithlowlatency.PeeringlinksthatconnectCPsdirectlytoaccessInternetserviceproviders(accessISPs)havebeenusedforthispurposethusprovidingone-hopASpathsfromCPstousers.Whileprovidingimprovedlatency,thesepeeringlinksstilldonotgiveCPscontrolovertheentireend-to-endpathtotheirusers.ThishasmadeitdifficultforCPstocompletelymanageuserexperience.Motivatedbythis,weproposethedeploymentofZero-HopNetworks(ZeN),whereaCP'sentireend-to-endpathtousersisunderitscontrol.WebelieveitisimportanttorespondtothecompellingdemandforZeNandenableitsprovisionoverthesharedInternetinfrastructuresothatallmaycontinuetoreapitsbenefits.InthispaperwelayoutthevisionforZeN,describingitsgoalsandchallenges.WeproposetodeployZeNbyallowingCPstoextendtheirnetwork'scontrolovertheaccessISPsubstrateinawaythatallowstheCPtocontroltheentireend-to-endpath.WedeveloptwostrawmanarchitecturesbasedonSoftware-DefinedNetworkingideas:onebasedonresourcereservationandtheotherbasedonnetworkvirtualization.WealsodiscusssomeelementsofaresearchagendathatisneededtobringZeNdeploymentstorealization.
StructuredOverlayNetworksforaNewGenerationofInternetServices AmyBabay(JohnsHopkinsUniversity),ClaudiuDanilov(BoeingResearchandTechnology),JohnLane(LTNGLobalCommunications),MichalMiskin-Amir(LTNGlobalCommunications,SpreadConceptsLLC),DanielObenshain(JohnsHopkinsUniversity),JohnSchultz(LTNGlobalCommunications,SpreadConceptsLLC),JonathanStanton(LTNGlobalCommunications,SpreadConceptsLLC),ThomasTantillo(JohnsHopkinsUniversity),YairAmir(JohnsHopkinsUniversity,LTNGlobalCommunications,SpreadConceptsLLC)
ThedramaticsuccessandscalingoftheInternetwasmadepossiblebythecoreprincipleofkeepingitsimpleinthemiddleandsmartattheedge(ortheend-to-endprinciple).However,newapplicationsbringnewdemands,andformanyemergingapplications,theInternetparadigmpresentslimitations.ForapplicationsinthisnewgenerationofInternetservices,structuredoverlaynetworksofferapowerfulframeworkfordeployingspecializedprotocolsthatcanprovidenewcapabilitiesbeyondwhattheInternetnativelysupportsbyleveragingglobalstateandin-networkprocessing.Thestructuredoverlayconceptincludesthreeprinciples:Aresilientnetworkarchitecture,aflexibleoverlaynodesoftwarearchitecturethatexploitsglobalstateandunlimitedprogrammability,andflow-basedprocessing.Wedemonstratetheeffectivenessofstructuredoverlaynetworksinsupportingtodaysdemandingapplicationsandproposeforward-lookingideasforleveragingtheframeworktodevelopprotocolsthatpushtheboundariesofwhatispossibleintermsofperformanceandresilience.
EnsuringNetworkNeutralityforFutureDistributedSystems ThiagoGarrett(FederalUniversityofParana),SchahramDustdar(TUWien),LuisC.E.Bona(FederalUniversityofParana),EliasP.DuarteJr.(FederalUniversityofParana)
NetworkNeutralityisessentialforensuringalevelplayingfieldforthedevelopmentofnewapplicationsandservicesontheInternet.Lawsandrulesalonemightnotbeenoughtoprotectinnovation,faircompetitionandconsumersfreedomofchoiceonline.TheresearchcommunityhastheresponsibilitytoproposesolutionsthatrevealdiscriminatorytrafficmanagementmechanismsontheInternet.Wepresentthepotentialrisksofanon-neutralInternet,identifyseveralopenchallengesfordesigningsolutionsthatdetecttrafficdifferentiation,andproposeamodelthataddressessuchchallengesbytakingadvantageofdistributedsystemstechnologies.
UncoveringtheUsefulStructuresofComplexNetworksinSocially-RichandDynamicEnvironments JieWu(TempleUniversity)
Manygroupactivitiescanberepresentedasacomplexnetworkwhereentities(vertices)areconnectedinpairsbylines(edges).Uncoveringausefulglobalstructureofcomplexnetworksisimportantforunderstandingsystembehaviorsandinprovidingglobalguidanceforapplicationdesigns.Webrieflyreviewexistingnetworkmodels,discussseveraltoolsusedinthetraditionalgraphtheory,distributedcomputing,distributedsystems,andsocialnetworkcommunities,andpointouttheirlimitations.Wediscussopportunitiestouncoverthestructuralpropertiesofcomplexnetworks,especiallyinamobileenvironment,andwesummarizethreepromisingapproachesforuncoveringusefulstructures:trimming,layering,andremapping.Finally,wepresentsomechallengesinalgorithmictechniques,withafocusondistributedandlocalizedsolutions,torepresentvariousstructures.
FutureNetworkingChallenges:TheCaseofMobileAugmentedReality TristanBraud(TheHongKongUniversityofScienceandTechnology),FarshidHassaniBijarbooneh(TheHongKongUniversityofScienceandTechnology),DimitrisChatzopoulos(TheHongKongUniversityofScienceandTechnology),PanHui(TheHongKongUniversityofScienceandTechnology)
Mobileaugmentedreality(MAR)applicationsaregainingpopularityduetothewideadoptionofmobileandespeciallywearabledevices.SuchdevicesoftenpresentlimitedhardwarecapabilitieswhileMARapplicationsoftenrelyoncomputationallyintensivecomputervisionalgorithmswithextremelatencyrequirements.Tocompensateforthelackofcomputingpower,offloadingdataprocessingtoadistantmachineisoftendesired.However,thisprocessintroducesnewconstrainsintheapplication,especiallyintermsoflatencyandbandwidth.Ifcurrentnetworkinfrastructuresarenotreadyforsuchtraffic,weenvisionthatfuturewirelessnetworkssuchas5GwillrapidlybesaturatedbyresourcehungryMARapplications.Moreover,duetothehighvarianceofwirelessnetworks,MARapplicationsshouldnotrelyonlyontheevolutionofinfrastructures.Inthisarticle,weanalyzeMARapplicationsandjustifytheirneedforaccessingexternalinfrastructure.Afterareviewoftheexistingnetworkinfrastructuresandprotocols,wedefineguidelinesforfuturereal-timeandmultimediatransportprotocols,withafocusonMARoffloading.
SoftwareDefinedCyberinfrastructure IanFoster(ArgonneNationalLaboratoryandTheUniversityofChicago),BenBlaiszik(TheUniversityofChicago),KyleChard(ComputationInstitute,UniversityofChicagoandArgonneNationalLab),RyanChard(VictoriaUniversityofWellington)
![Page 90: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/90.jpg)
90
Withinandacrossthousandsofsciencelabs,researchersandstudentsstruggletomanagedataproducedinexperiments,simulations,andanalyses.Largelymanualresearchdatalifecyclemanagementprocessesmeanthatmuchtimeiswasted,researchresultsareoftenirreproducible,anddatasharingandreuseremainrare.Inresponse,weproposeanewapproachtodatalifecyclemanagementinwhichresearchersareempoweredtodefinetheactionstobeperformedatindividualstoragesystemswhendataarecreatedormodified:actionssuchasanalysis,transformation,copying,andpublication.Wetermthisapproachsoftware-definedcyberinfrastructurebecauseuserscanimplementpowerfuldatamanagementpoliciesbydeployingrulestolocalstoragesystems,muchassoftware-definednetworkingallowsuserstoconfigurenetworksbydeployingrulestoswitches.Wearguethatthisapproachcanenableanewclassofresponsivedistributedstorageinfrastructurethatwillaccelerateresearchinnovationbyallowinganyresearchertoassociatedataworkflowswithdatasources,whetherlocalorremote,forsuchpurposesasdataingest,characterization,indexing,andsharing.Wereportonearlyexperimentswiththisapproachinthecontextofexperimentalscience,inwhichasimpleif-trigger-then-action(IFTA)notationisusedtodefinerules.
Vision3:NextGenerationCloudandEdgeServicesComputingintheContinuum:CombiningPervasiveDevicesandServicestoSupportData-drivenApplications ManishParashar(RutgersUniversity),MoustafaAbdelbaky(RutgersUniversity),MengsongZou(RutgersUniversity),AliRezaZamani(RutgersUniversity),EduardRenart(RutgersUniversity),JavierDiaz-Montes(RutgersUniversity)
Theexponentialgrowthofdigitaldatasourceshas,thepotentialtotransformallaspectsofsocietyandourlives.,However,toachievethisimpactthedatahastobeprocessed,inatimelymannertoextractcriticalinsightsthatcandrive,decisionmaking.Further,traditionalapproachesthatrelyon,movingdatatoremotedatacentersforprocessingarenolonger,feasible.Instead,newapproachesthateffectivelyleveragedistributed,computationalinfrastructureandservicesarenecessary.,Specifically,theseapproachesmustseamlesslycombineresources,andservicesattheedge,inthecore,andalongthedata,pathasneeded.Thispaperpresentsourvisionforenabling,anapproachforcomputinginthecontinuum,i.e.,realizinga,fluidecosystemwheredistributedresourcesandservicesare,programmaticallyaggregatedon-demandtosupportemerging,data-drivenapplicationworkflows.Thisvisioncallsfornovel,solutionsforfederatinginfrastructure,programmingapplications,andservices,andcomposingdynamicworkflows,whichare,capableofreactinginreal-timetounpredictabledatasizes,,availabilities,locations,andrates.
Decision-drivenExecution:ADistributedResourceManagementParadigmfortheAgeofIoT TarekAbdelzaher(UIUC),TanvirAlAmin(UIUC),AmotzBar-Noy(UIUC),WilliamDron(BBN),RameshGovindan(USC),ReginaldHobbs(ARL),ShaohanHu(IBM),Jung-EunKim(UIUC),ShuochaoYao(UIUC),YiranZhao(UIUC)
Thispaperintroducesanovelparadigmforresourcemanagementindistributedsystems,calleddecision-drivenexecution.Theparadigmisappropriateformission-drivensystems,wherethegoalistoenablefaster,leaner,andmoreeffectivedecisionmaking.Allresourceconsumption,inthisparadigm,istiedtotheneedsofmakingdecisionsonalternativecoursesofaction.Apointofdeparturefromtraditionalarchitecturesliesininterfacesthatallowapplicationstospecifytheirunderlyingdecisionlogic.Thisspecification,inturn,allowsthesystemtoreasonaboutmosteffectivemeanstomeetinformationneedsofdecisions,resultinginsimulataneousoptimizationofdecisionaccuracy,cost,andspeed.Thepaperdiscussestheoverallvisionofdecision-drivenexecution,outliningpreliminaryworkandnovelchallenges.
ACTiCLOUD:EnablingtheNextGenerationofCloudApplications GeorgiosGoumas(NationalTechnicalUniversityofAthens),KonstantinosNikas(ComputingSystemsLaboratory,NTUA),EwnetuBayuhLakew(Dept.ofComputingScience,UmeaUniversity),ChristosKotselidis(TheUniversityofManchester),VasileiosKarakostas(ComputingSystemsLaboratory,NTUA),AtleVesterkjaer(Numascale),EinarRustad(Numascale),JohnGoodacre(Kaleao),AndrewAttwood(Kaleao),MichailFlouris(OnApp),JohnThomson(OnApp),NikosFoutris(TheUniversityofManchester),MikelLujan(TheUniversityofManchester),YingZhang(MonetDBSolutions),PanagiotisKoutsourakis(MonetDBSolutions),MartinKersten(MonetDBSolutions),JimWebber(NeoTechnology),DavideGrohmann(NeoTechnology),ErikElmroth(Dept.ofComputingScience,UmeaUniversity),LuisTomas(Dept.ofComputingScience,UmeaUniversity),NectariosKoziris(NationalTechnicalUniversityofAthens)
Despitetheirproliferationasadominantcomputingparadigm,cloudcomputingsystemslackeffectivemechanismstomanagetheirvastamountsofresourcesefficiently.Resourcesarestrandedandfragmented,ultimatelylimitingcloudsystemsapplicabilitytolargeclassesofcriticalapplicationsthatposenon-moderateresourcedemands.Eliminatingcurrenttechnologicalbarriersofactualfluidityandscalabilityofcloudresourcesisessentialtostrengthencloudcomputingsroleasacriticalcornerstoneforthedigitaleconomy.ACTiCLOUDproposesanovelcloudarchitecturethatbreakstheexistingscale-upandshare-nothingbarriersandenablestheholisticmanagementofphysicalresourcesbothatthelocalcloudsiteandatdistributedlevels.Specifically,itmakesadvancementsinthecloudresourcemanagementstacksbyextendingstate-of-the-arthypervisortechnologybeyondthephysicalserverboundaryandlocalizedcloudmanagementsystemtoprovideanholisticresourcemanagementwithinarack,withinasite,andacrossdistributedcloudsites.Ontopofthis,ACTiCLOUDwilladaptandoptimizesystemlibrariesandruntimes(e.g.,JVM)aswellasACTiCLOUD-nativeapplications,whichareextremelydemanding,andcriticalclassesofapplicationsthatcurrentlyfaceseveredifficultiesinmatchingtheirresourcerequirementstostate-of-the-artcloudofferings.
JointCloud:ACross-CloudCooperationArchitectureforIntegratedInternetServiceCustomization HuaiminWang(NationalUniversityofDefenseTechnology),PeichangShi(NationalUniversityofDefenseTechnology),YimingZhang(NationalUniversityofDefenseTechnology)
CloudcomputinghascompletelychangedtheeconomicsofITindustry.Recently,thenewformofsharedglobaleconomyrequirescloudservicestobecollaborativelyprovisionedbydifferentcloudprovidersinaGeo-distributedmanner,whichbringsseverechallengesinserviceperformanceandcost.Toaddressthisproblem,inthispaperweproposeJointCloud,across-cloudcooperationarchitectureforintegratedInternetservicecustomization.JointCloudborrowstheideafromairlinealliancesandaimsatempoweringthecooperationamongmultiplecloudstoprovideefficientcross-cloudservices.JointCloudfocusesnotonlyontheverticalintegrationofcloudresourcesbutalsoonthehorizontalcooperationamongdifferentcloudvendors.ThispaperdescribestheconceptandarchitectureofJointCloud,aswellastheinitialdesignsofJointCloudskeycomponents,namely,communication,storage,andcomputation.
SupportingDataAnalyticsApplicationsWhichUtilizeCognitiveServices ArunIyengar(IBMResearch)
![Page 91: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/91.jpg)
91
AwidevarietyofservicesareavailableovertheWebwhichcandramaticallyimprovethefunctionalityofapplications.Theseservicesincludeinformationretrieval(includingdatalookupsfromavarietyofsourcesandWebsearches),naturallanguageunderstanding,visualrecognition,anddatastorage.Akeyproblemishowtoprovidesupportforapplicationswhichusetheseservices.Thispaperpresentsarichsoftwaredevelopmentkit(SDK)whichaccessestheseservicesandprovidesavarietyoffeaturesapplicationsneedtousetheseservices,optimizeperformance,andcomparethem.AkeyaspectofourSDKisitssupportfornaturallanguageunderstandingservices.WealsopresentapersonalizedknowledgebasebuiltontopofourrichSDKthatusespublicallyavailabledatasourcesaswellasprivateinformation.Theknowledgebasesupportsdataanalysisandreasoningoverdata.
TrillionOperationsKey-ValueStorageEngine:RevisitingtheMissionCriticalAnalyticsStorageSoftwareStack SangeethaSeshadri(IBMAlmadenResearchCenter),LawrenceChiu(IBMAlmadenResearchCenter),PaulMuench(IBMAlmadenResearchCenter)
Dataisthenewnaturalresourceofthiscentury.Asdatavolumesgrowandapplicationsaimedatmonetizingthedatacontinuetoevolve,dataprocessingplatformsareexpectedtomeetnewscale,performance,reliabilityanddataretentionrequirements.Atthesametime,storagehardwarecontinuestoimproveinperformanceandprice-performance.Inthispaper,wepresentTOKVS-TrillionOperationKey-ValueStore,aNoSQLstorageenginethatredefinesthestoragesoftwarestacktomeettherequirementsofnext-generationapplicationsonnext-generationhardware.
Vision4:SecurityandTrustinFutureSystemsHowComputerScienceRiskstoLoseItsInnocence,andShouldAttempttoTakeResponsibilityKarlAberer(EPFL)
Computerscienceisplayingadrivingroleintransformingtoday’ssocietythroughinformationtechnology.Inthistransformationweobservepowershiftsincreasinglystrengtheningcentralisedorganisations,whicharenegativelyperceivedbymanypeople.Weoutlinetechnicalquestionsthatcomputerscienceshouldpayattentiontoinordertoenableindividualsinpreservingtheirinterestandtotakemeaningfuldecisionsbasedonreliableinformation.
ACognitivePolicyFrameworkforNext-GenerationDistributedFederatedSystems-ConceptsandResearchDirections ElisaBertino(PurdueUniversity),SeraphinCalo(IBM),MarounTouma(IBM),DineshVerma(IBM),ChristopherWilliams(UKDSTL),BrianRivera(ArmyResearchLabs)
Next-generationcollaborativeactivitiesandmissionswillbecarriedoutbyautonomousgroupsofdeviceswithalargevarietyofcognitivecapabilities.Thesedeviceswillhavetooperateinenvironmentscharacterizedbyuncertainty,insecurity(bothphysicalandcyber),andinstability.Insuchenvironments,communicationsmaybefragmented.Properpolicy-basedmanagementofsuchautonomousdevicegroupsisthuscritical.Howevercurrentpolicymanagementsystemshavemanylimitations,includinglackofflexibility.Inthispaper,wearticulatenovelarchitecturalapproachesaddressingtherequirementsfortheeffectivemanagementofautonomousgroupsofdevicesanddiscussthenotionofgenerativepoliciesanovelparadigmthatenhancestheflexibilityofpolicy-basedapproachestomanagement.Inthispaper,wealsosurveytypesofpolicythatareessentialformanagingdevicegroups.Eventhoughmanysuchpolicytypesexistinconventionalsettings,theiruseinourcontextposesnovelchallengesthatwearticulateinthepaper.Wealsointroducearesearchroadmapdiscussingseveralresearchdirectionstowardsthedevelopmentofacognitiveandflexiblepolicy-basedapproachtothemanagementofautonomousgroupsofdevicesforcollaborativemissions.Finally,asourproposedpolicyparadigmisdata-intensive,wediscusstheproblemofsupplyingthedatarequiredforpolicydecisionsinenvironmentscharacterizedbymobility,uncertainly,andfragmentedcommunications.
MachinetoMachineTrustinSmartCities MargaretLoper(GeorgiaTechResearchInstitute),BrianSwenson(GeorgiaInstituteofTechnology)
Inthecomingdecades,wewillliveinaworldsurroundedbytensofbillionsofdevicesthatwillinteroperateandcollaborateinanefforttodeliverpersonalizedandautonomicservices.ThisparadigmofsmartobjectsandsmartthingsinterconnectedandubiquitouslysurroundingusiscalledtheInternetofThings(IoT).CitiesmaybethefirsttobenefitfromtheIoT,butrelianceonthesemachinestomakedecisionshasprofoundimplicationsfortrust,andmakesmechanismsforexpressingandreasoningabouttrustessential.ThispaperintroducestheprojectfundedbytheGeorgiaTechResearchInstitutetolookatseveraldimensionsofMachinetoMachineTrustinthecontextofSmartCities.
LateralThinkingforTrustworthyApps HermannHärtig(TechnischeUniversitätDresden),MichaelRoitzsch(TechnischeUniversitätDresden),CarstenWeinhold(TechnischeUniversitätDresden),AdamLackorzynski(TechnischeUniversitätDresden)
Thegrowingcomputerizationofcriticalinfrastructureaswellasthepervasivenessofcomputingineverydaylifehasledtoincreasedinterestinsecureapplicationdevelopment.ExemplifiedbyARMTrustZoneandIntelSGX,weobserveaflurryofnewsecuritytechnologies,butalackofanarchitecturalvision.Weareconvincedthatpointsolutionsarenotsufficienttoaddresstheoverallchallengeofsecuresystemdesign.Inthispaper,wesketchourtakeonatrustedcomponentecosystemofsmallindividualbuildingblockswithstrongisolation.Inourview,applicationsshouldnolongerbedesignedasmassivestacksofverticallylayeredframeworks,butinsteadashorizontalaggregatesofmutuallyisolatedcomponentsthatcollaborateacrossmachineboundariestoprovideaservice.Lateralthinkingisneededtocreatesecuresystemsgoingforward.
RumorInitiatorDetectioninInfectedSignedNetworks JiaweiZhang(UniversityofIllinoisatChicago),CharuC.Aggarwal(IBMT.J.WatsonResearchCenter),PhilipS.Yu(UniversityofIllinoisatChicago)
Inmanycases,theinformationspreadinanonlinenetworkmaynotalwaysbetruthfulorcorrect;suchinformationcorrespondstorumors.Inrecentyears,signednetworkshavebecomeincreasinglypopularbecauseoftheirabilitytorepresentdiverserelationshipssuchasfriends,enemies,trust,anddistrust.Signednetworksareidealforinformationflowinanetworkwithvaryingbeliefs(trustordistrust)aboutfacts.Inthispaper,wewillstudytheproblemofinfluenceanalysisanddiffusionmodelsinsignednetworksandinvestigatetheproblemofrumorinitiatordetection,giventhestateofthenetworkatagivenmomentintime.Conventionalinformationdiffusionmodelsforunsignednetworkscannotbeappliedtosignednetworksdirectly,andweshowthattherumorinitiatordetectionproblemisNP-hard.Weproposeanewinformationdiffusionmodel,referredtoasasyMmetricFlippingCascade(MFC),tomodelthepropagationofinformationinsignednetworks.BasedonMFC,anovelframework,RumorInitiatorDetector(RID),isintroducedtodeterminethepotentialnumberandtheidentityoftherumorinitiatorsfromthestateofthenetworkatagiventime.Extensiveexperimentsconductedonreal-worldsignednetworksdemonstratethatMFCworksverywellinmodelinginformationdiffusioninsignednetworksandRIDcansignificantlyoutperformothercomparisonmethodsinidentifyingrumorinitiators.
![Page 92: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/92.jpg)
92
AddressingSmartphone-basedMulti-factorAuthenticationviaHardware-rootedTechnologiesZhongjieBa(TheStateUniversityofNewYorkatBuffalo),KuiRen(TheStateUniversityofNewYorkatBuffalo)
Multi-factorauthenticationisawell-recognizedaccesscontrolmethodthatenhancesthesecurityofuserssensitivedataandidentities.Asuccessfulauthenticationattemptrequiresausertocorrectlypresenttwoormoreauthenticationfactorssuchasknowledgefactors,possessionfactorsandinherencefactors.Forsmartphone-basedmulti-factorauthentication,apromisingwaytoauthenticateauseristoverifyhispossessionofalegitimatesmartphone,whichcallsforsecureandusabledeviceauthenticationschemes.Inthisarticle,weproposetoauthenticateadevicethroughtrackingthehardwarefingerprintofitsbuilt-insensor.Wefirstreviewtheexistinghardwarerootedidentificationmethodsanddiscussthemeritsofapplyingahardwarefingerprintasasmartphonesuniqueidentity.Then,weanalyzethesecurityissuesunderlyingthesemethodsandidentifytwosecurityrequirementsfortheidentificationmethodstobeusedinanauthenticationscheme:FingerprintLeakageResilienceandFingerprintForgeryResilience.Finally,welookintoaspecifichardwarefingerprintoriginallyusedfordigitalcameras.Weanalyzethefeasibilityofapplyingthisfingerprinttodifferentiateoff-the-shelfsmartphonesandlistseveralchallengingpracticalissuesunderlyingthismethod.
Vision5:FutureDistributedSystemsEnablingwideareadataanalyticswithCollaborativeDistributedProcessingPipes(CDPPs) AnjaFeldmann(TUBerlin),ManfredHauswirth(TUBerlin),VolkerMarkl(TUBerlin)
TheMillibottleneckTheoryofPerformanceBugs,andItsExperimentalVerification CaltonPu(GeorgiaInstituteofTechnology),JoshuaKimball(GeorgiaInstituteofTechnology),Chien-AnLai(GeorgiaInstituteofTechnology),TaoZhu(GeorgiaInstituteofTechnology),JackLi,JunheePark,QingyangWang,DeepalJayasinghe,PengchengXiong,SimonMalkowski,QinyiWu,GueyoungJung,YounggyunKoh,GalenSwint
Theperformanceofn-tierweb-facingapplicationsoftensufferfromresponsetimelong-tailproblem.Withrelativelylowresourceutilization(lessthan50%)andthevastmajorityofrequestsreturningwithinafewmilliseconds,anon-negligiblenumberofnormallyshortrequestsmaytakesecondstoreturn.Weproposethemillibottlenecktheoryofperformancebugs(thatleadtolong-tailproblems).Severalcasestudieshaveconfirmedthemillibottlenecks(thatlastafewtenstohundredsofmilliseconds)ascausalagentsoflongrequests.Aconcreteexample(garbagecollection)illustratestheexperimentalverificationofmillibottlenecks.Anopensourcefine-grainmonitoringtoolkitisbeingdevelopedtofacilitatetheexperimentalresearchonmillibottlenecks.
Exacution:EnhancingScientificDataManagementforExascale ScottKlasky(OakRidgeNationalLaboratory),EricSuchyta(OakRidgeNationalLab),MarkAinsworth(BrownUniversity),QingLiu(NewJerseyInstituteofTechnology),BenWhitney(BrownUniversity),MatthewWolf(OakRidgeNationalLaboratory),JongChoi(OakRidgeNationalLaboratory),IanFoster(ArgonneNationalLaboratory),MarkKim(OakRidgeNationalLaboratory),JeremyLogan(UniversityOfTennesseeKnoxville),KshitijMehta(OakRidgeNationalLaboratory),ToddMunson(ArgonneNationalLaboratory),GeorgeOstrouchov(OakRidgeNationalLaboratory),ManishParashar(RutgersUniversity),NorbertPodhorszk(OakRidgeNationalLaboratory),DavidPugmire(OakRidgeNationalLaboratory),LipengWan(OakRidgeNationalLaboratory)
Aswecontinuetowardexascale,scientificdatavolumeiscontinuingtoscaleandbecomingmoreburdensometomanage.Inthispaper,welayoutopportunitiestoenhancestateoftheartdatamanagementtechniques.Weemphasizewell-principleddatacompression,andusingittoachieveprogressiverefinement.ThiscanbothaccelerateI/Oandaffordtheuserincreasedflexibilitywhensheinteractswiththedata.Theformulationnaturallymapsontoenablingonetopartitiontheprogressivelyimprovingqualityrepresentationsofthesamedataquantityintodifferentmedia-typedestinations,tokeepthehighestpriorityinformationascloseaspossibletothecomputation,andtakeadvantageofdeepeningmemory/storagehierarchiesinwaysnotpreviouslypossible.Carefulmonitoringisrequisitetoourvision,notonlytoverifythatcompressionhasnoteliminatedsalientfeaturesinthedata,butalsotobetterunderstandtheperformanceofhighperformancescientificapplications.Increasedmathematicalrigorwouldideal,tohelpbringcompressiononabetter-understoodtheoreticalfooting,closertotherelevantscientifictheory,moreawareofconstraintsimposedbythescience,andmoretightlyerrorcontrolled.Throughout,wehighlightpathfindingresearchwehavebegunexploringrelatedthesetopics,andcommenttowardfutureworkthatwillbeneeded.
HardwareAccelerationLandscapeforDistributedReal-timeAnalytics:VirtuesandLimitations MohammadrezaNajafi(TechnischeUniversitatMunchen),KaiwenZhang(TechnischeUniversitatMunchen),Hans-ArnoJacobsen(UniversityofToronto),MohammadSadoghi(PurdueUniversity)
Arguably,wearenowwitnessinganewtechnologicalrevolutionwiththepotentialthatrangesfromtransformingourday-to-daylifeexperiences(e.g.,personalizedmedicineandeducation)totransformingeverysingleindustry(e.g.,data-drivenhealthcare,commerce,agriculture,andmining).Atthecoreofthisrevolutionliesdata.Thistransformationisfacilitatedbysensing,gathering,andconnectingallphysicalentitiestoconstructarichanddynamiccomputationalmodelofrealityinreal-time.Everyprocedureandeverydecisionneededinthephysicalworldwillsoonbeoptimizedinreal-timebyingestingandanalyzingmassivevolumeofpresentandpastdataatanunprecedentedvelocity.Tocopewithsuchextremescale,wearguetheneedtorevisitthehardwareandsoftwareco-designinlightoftwokeytechnologicaladvancements.Firstisthevirtualizationofcomputationandstorageoverhighlydistributeddatacentersspanningacrosscontinents.Secondistheemergenceofavarietyofspecializedhardwareacceleratorsthatcomplementthetraditionalgeneral-purposeprocessors.Wearguethereisanimminentneedtoexploitandunifythesetwotrendsinordertounleashandharnessthepowerofdatainreal-time.Inthispaper,wefocusonpresentingaformulationandcharacterizationofhardwareaccelerationlandscapegearedtowardsreal-timeanalyticsinthecloudenvironment.Ourgoalistoassistbothresearchersandpractitionersnavigatingthenewlyrevivedfieldofsoftwareandhardwareco-designforbuildingnextgenerationdistributedsystems.Wefurtherpresentacasestudytoexploresoftwareandhardwareinterplayfordesigningdistributedrealtimestreamprocessing.
CoordinatingDistributedSpeakingObjects MarcoLippi(DISMI–UniversitàdiModenaeReggioEmilia),MarcoMamei(DISMI–UniversitàdiModenaeReggioEmilia),StefanoMariani(DISMI–UniversitàdiModenaeReggioEmilia),FrancoZambonelli(DISMI–UniversitàdiModenaeReggioEmilia)
![Page 93: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/93.jpg)
93
Inthispaperwesketchavisionoffutureenvironmentsdenselypopulatedbysmartsensorsandactuatorspossiblyembeddedineverydayobjectsthat,ratherthansimplyproducingstreamsofdata,arecapableofunderstandingandreporting,viafactualassertionsandarguments,aboutwhatishappening(forsensors)andaboutwhattheycanmakepossiblyhappen(foractuators).Thesespeakingobjectsformthenodesofadensedistributedcomputinginfrastructurethatcanbeexploitedtomonitorandcontrolactivitiesinoureverydayenvironment.However,thenatureofspeakingobjectswilldramaticallychangetheapproachestoimplementingandcoordinatingtheactivitiesofdistributedprocesses.Infact,distributedcoordinationislikelytobecomeassociatedwiththecapabilityofargumentingaboutsituationsandaboutthecurrentstateoftheaffairs,withtheaimoftriggeringanddirectingproperdistributedconversationstocollectivelyreachafuturedesirablestateoftheaffairs.Inthispaperwediscusshowsuchanovelvisioncanbuilduponsomereadilyavailabletechnologies,andwehighlighttheresearchchallengesthatitposes.Twocasestudiesareusedthroughoutthepaperasexemplaryscenarios.
Model-DrivenDomain-SpecificMiddleware FabioCosta(FederalUniversityofGoias),KarlMorris(TempleUniversity),FabioKon(UniversityofSãoPaulo),PeterClarke(FloridaInternationalUniversity)
Middlewarewasintroducedtofacilitatethedevelopmentofsophisticatedapplicationsbasedonauniformmethodologyandindustrystandards.However,earlyresearchandpracticesuggestedthatnoone-size-fits-allapproachwassuitableforallapplicationdomainsandscenarios.Thisgaverisetoindustryinitiativestostandardizedomain-specificmiddlewareservicesandprofiles,aswellasresearcheffortsonconfigurable,reflective,andadaptivemiddleware.Theindustrysapproachledtoeasydeployment,althoughwithalevelofflexibilitylimitedbytheextentofexistingprofiles.Theapproachoftheresearchcommunity,ontheotherhand,enabledhighflexibility,allowinganymiddlewareconfigurationtobedefined.Nevertheless,creatingsoundconfigurationsusingthisapproachisachallengingtask,limitingthetargetaudiencetoexpertengineers.Asaconsequence,bothinitiativesdonotscalewiththecurrentproliferationofspecializedapplicationdomains.Inthispaper,wetargetthisproblemwithanapproachthatleveragesmodel-drivenengineeringfortheconstructionofdomain-specificmiddlewareplatforms.Asetofhigh-level,yetexpressive,buildingblocksisdefinedintheformofametamodel,whichisusedtocreatemodelsthatspecifythedesiredmiddlewareconfiguration.Wearguethatthisapproachenablestherapiddevelopmentofmiddlewareplatformstomatchtheproliferationofapplicationdomains,atthesametimeasitdoesnotrequireper-applicationmiddlewareconstructionorevenhighlyskilledmiddlewareengineers.Wepresentthecurrentstateofourresearchanddiscussresearchdirectionstofullyrealizetheapproach.
Vision6:InnovationinBigDataSystemsOntheDesignofaBlockchainPlatformforClinicalTrialandPrecisionMedicine ZonyinShae(ASIAUniversity,Taiwan),JeffreyTsai(ASIAUniversity,Taiwan)
Thispaperproposesablockchainplatformarchitectureforclinicaltrialandprecisionmedicineanddiscussesvariousdesignaspectsandprovidessomeinsightsinthetechnologyrequirementsandchallenges.Weidentify4newsystemarchitecturecomponentsthatarerequiredtobebuiltontopoftraditionalblockchainanddiscusstheirtechnologychallengesinourblockchainplatform:(a)anewblockchainbasedgeneraldistributedandparallelcomputingparadigmcomponenttodeviseandstudyparallelcomputingmethodologyforbigdataanalytics,(b)blockchainapplicationdatamanagementcomponentfordataintegrity,bigdataintegration,andintegratingdisparityofmedicalrelateddata,(c)verifiableanonymousidentitymanagementcomponentforidentityprivacyforbothpersonandInternetofThings(IoT)devicesandsecuredataaccesstomakepossibleofthepatientcentricmedicine,and(d)trustdatasharingmanagementcomponenttoenableatrustmedicaldataecosystemforcollaborativeresearch.
TowardsDataflow-basedGraphAccelerator HaiJin(HuazhongUniversityofScienceandTechnology),PengchengYao(HuazhongUniversityofScienceandTechnology),XiaofeiLiao(HuazhongUniversityofScienceandTechnology),LongZheng(HuazhongUniversityofScienceandTechnology),XianliangLi(HuazhongUniversityofScienceandTechnology)
Existinggraphprocessingframeworksgreatlyimprovetheperformanceofmemorysubsystem,buttheyarestillsubjecttotheunderlyingmodernprocessor,resultinginthepotentialinefficienciesforgraphprocessinginthesenseoflowinstructionlevelparallelismandhighbranchmisprediction.Theseinefficiencies,inaccordancewithourcomprehensivemicro-architecturalstudy,mainlyariseoutofawealthofdatadependencies,serialsemanticofinstructionstreams,andcomplexconditionalinstructionsingraphprocessing.Inthispaper,weproposethatafundamentalshiftofapproachisnecessarytobreakthroughtheinefficienciesoftheunderlyingprocessorviathedataflowparadigm.Itisverifiedthattheideaofapplyingdataflowapproachintographprocessingisextremelyappealingforthefollowingtworeasons.First,astheexecutionandretirementofinstructionsonlydependontheavailabilityofinputdataindataflowmodel,ahighdegreeofparallelismcanbethereforeprovidedtorelaxtheheavydependencyandserialsemantic.Second,dataflowisguaranteedtomakeitpossibletoreducethecostsofbranchmispredictionbysimultaneouslyexecutingallbranchesofaconditionalinstruction.Consequently,wemakethepreliminaryattempttodevelopthedataflowinsightintoaspecializedgraphaccelerator.Webelievethatourworkwouldopenawiderangeofopportunitiestoimproveperformanceofcomputationandmemoryaccessforlarge-scalegraphprocessing.
TowardsaRISCFrameworkforEfficientContextualizationinIoT DimitriosGeorgakopoulos(SwinburneUniversity),AliYavari(RMITUniversity),PremPrakashJayaraman(SwinburneUniversity),RajivRanjan(NewcastleUniversity)
TheInternetofThings(IoT)isanewinternetevolutionthatinvolvesconnectingbillionsofinternet-connecteddeviceswerefertoasIoTthings.ThesedevicescancommunicatedirectlyandintelligentlyovertheInternet,andgenerateamassiveamountofdatathatneedstobebyavarietyofIoTapplications.ThispaperfocusesontheautomaticcontextualisationofIoTdata,whichalsoinvolvesdistillinginformationandknowledgefromIoTaimingtosimplifyansweringthefollowingfundamentalquestionsthatoftenariseinIoTapplications:WhichdatacollectedbyIoTarerelevanttomyselfandtheIoTThingsIcarefor?Relatedworkaroundcontextmanagementandcontextualisationrangesfromdatabasetechniquesthatinvolvequeryre-writing,tosemanticwebandrule-basedcontextmanagementapproaches,tomachinelearninganddatascience-basedsolutionsinmobileandambientcomputing.Allsuchexistingapproacheshavetwomainaspectsincommon:Theyarehighlyincompatibleandhorriblyinefficientfromascalabilityandperformanceperspective.Inthispaper,wediscussanewRISCContextualisationFramework(RCF)wehavedeveloped,implementedkeyaspectoff,andassessesitsscalability.RCFprovidesfundamentalcontextualisationconceptsthatcanbemappedtoallexistingcontextualisationapproachesforIoTdata(andinthissense,itprovidesacommondenominatorthatunifiesthecontextualisationspace).RCFcanbeeasilyimplementedasacloud-basedservice,andprovidesbetterscalabilityandperformancethatanyoftheexistingcontentmanagementandcontextualisationapproachintheIoTspace.
![Page 94: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/94.jpg)
94
TheFutureoftheSemanticWeb:PrototypesonaGlobalDistributedFilesystem MichaelCochez(Fraunhofer-FIT),DominikHüser(RWTHAachenUniversity),StefanDecker(RWTHAachen)
Recently,prototypes(inthemeaningfamiliarfromprogramminglanguagessuchasJavascript)were(re-)introducedforknowledgerepresentationontheweb.However,thatworkhasaverytheoreticalfocusandamorepracticalsystemisindemand.Inthisvisionpaperwedescribehowadistributedfilesystemformsanaturalhabitatfortheprototypeknowledgerepresentation.Inparticular,wedescribehowweenvisiondeploymentofLinkedDataandPrototypeKnowledgebasesatopoftheInterPlanetaryFileSystem(IPFS),whichhasseveralusefulfeaturesmatchingtheneedsfortheprototypesystem.
OnBroadBigData SteffenStaab(InstitutWeST,UniversityKoblenz-LandauandWAIS,UniversityofSouthampton)
Abroaddatasystemexplicitlyrepresentsalargenumberofconceptsandpropertiestogetherwithitscorrespondingdataproper.Weobservethecharacteristicsofseveralbroaddatasystemsandelicitthreechallengeswewillneedtoresearchwhenscalingthesebroaddatasystemstobecomebigdatasystems,too.
![Page 95: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/95.jpg)
95
Short Paper Abstracts
ShortPaper1:DistributedOperatingSystems,Middleware,andAlgorithmsSRLB:ThePowerofChoicesinLoadBalancingwithSegmentRoutingYoannDesmouceaux(ÉcolePolytechnique),PierrePfister(CiscoSystems),JérômeTollet(CiscoSystems),MarkTownsley(CiscoSystems),ThomasClausen(ÉcolePolytechnique)
Networkload-balancersgenerallyeitherdonottakeapplicationstateintoaccount,ordosoatthecostofacentralizedmonitoringsystem.Thispaperintroducesaload-balancerrunningexclusivelywithintheIPforwardingplane,i.e.inanapplicationprotocolagnosticfashion–yetwhichstillprovidesapplication-awarenessandmakesreal-time,decentralizeddecisions.Tothatend,IPv6SegmentRoutingisusedtodirectdatapacketsfromanewflowthroughachainofcandidateservers,untilonedecidestoaccepttheconnection,basedonitslocalstate.Thisway,applicationsthemselvesnaturallydecideonhowtoshareincomingconnections,whileincurringminimalnetworkoverhead,andnoout-of-bandsignaling.Testsondifferentworkloads–includingrealisticworkloadssuchasreplayingactualWikipediaaccesstraffictowardsasetofreplicaWikipediainstances–showsignificantperformancebenefits,intermsofshorterresponsetimes,whencomparedtoatraditionalrandomload-balancer.
ImprovingEfficiencyofLinkClusteringonMulti-CoreMachines GuanhuaYan(BinghamtonUniversity)
Linkclusteringgroupsdifferentedgesinagraphaccordingtotheirsimilarities.Linkclusteringcanrevealtheoverlappingandhierarchicalorganizationsinawidespectrumofnetworks.Thisworkstudieshowtoimproveefficiencyoflinkclusteringalongthreedimensions,algorithm,modeling,andparallelization,onmulti-coremachines.Weevaluatetheefficiencyimprovedduetoeachofthethreedimensionsusingwordassociationgraphsextractedfromatwitterdataset.
S3:JointSchedulingandSourceSelectionforBackgroundTrafficinErasure-CodedStorage ShijingLi(GeorgeWashingtonUniversity),TianLan(GeorgeWashingtonUniversity),Moo-RyongRa(AT&TLabsResearch),RajeshPanta(AT&TLabsResearch)
Erasure-codedstoragesystemshavegainedconsiderableadoptionrecentlysincetheycanprovidethesamelevelofreliabilitywithsignificantlylowerstorageoverheadcomparedtoreplicatedsystems.However,backgroundtrafficofsuchsystems–e.g.repair,rebalance,backupandrecoverytraffic–oftenhaslargevolumeandconsumessignificantnetworkresources.Independentlyschedulingsuchtasksandselectingtheirsourcescaneasilycreateinterferenceamongdataflows,causingseveredeadlineviolation.Weshowthatthewell-knownheuristicschedulingalgorithmsfailtoconsiderimportantconstraints,thusresultinginunsatisfactoryperformance.Inthispaper,weclaimthatanoptimalschedulingalgorithmthataimstomaximizethenumberofbackgroundtaskscompletedbeforedeadlinesmustsimultaneouslyconsiderdeadline-awarescheduling,networktopology,chunkplacement,andtime-varyingresourceavailability.Tosolvethisproblem,weproposeanovelalgorithm,calledLinearProgrammingforSelectedTasks(LPST)tomaximizethenumberofsuccessfultasksandimproveoverallutilizationofthedatacenternetwork.ItjointlyschedulestasksandselectstheirsourcesbasedonanotionofRemainingTimeFlexibility,whichmeasurestheslacknessofthestartingtimeofatask.Weevaluatedtheefficacyofouralgorithmusingextensivesimulations.Ourresultsshowthat,undercertainscenarios,LPSTcanperform7x∼70xbetterthantheheuristicswhichblindlytreattheinfrastructureasacollectionofhomogeneousresources,and46.6%∼65.9%betterthanthealgorithmsthattakeintoaccountthenetworktopology.
OntheFeasibilityofInter-domainRoutingviaaSmallBrokerSet DongLin(HuaweiTechnologiesLtdCo.),DavidHui(HuaweiTechnologiesLtdCo.),WeijieWu(HuaweiTechnologiesLtdCo.),TingweiLiu(TheChineseUniversityofHongKong),YatingYang(BeijingInstituteofTechnology),YiWang(TsinghuaUniversity),JohnChi-ShingLui(ChineseUniversityofHongKong),GongZhang(HuaweiTechnologiesLtdCo.),YingtaoLi(HuaweiTechnologiesLtdCo.)
Thecurrentinter-domainroutingprotocol,namely,theBorderGatewayProtocol(BGP),cannotprovideend-to-end(E2E)quality-of-service(QoS)guarantees.Themainreasonisthatanautonomoussystem(AS)canonlyreceiveguaranteesfromitsfirsthopASesviaservicelevelagreements(SLAs).Butbeyondthefirsthop,QoSalongthepathfromsourcetodestinationASisnotwithinthesourceAS’scontrolregime.Inthispaper,weinvestigatethefeasibilityofprovidinghighQoS-guaranteedE2Etransitservicesbyutilizinga(small)setofASes/IXPstoserveas“brokers”toprovidesupervision,controlandresourcenegotiation.FindinganoptimalsetofASesasbrokerscanbeformulatedasaMaximumCoveragewithB-dominatingpathGuarantee(MCBG)problem,whichweprovetobeNP-hard.Toaddressthisproblem,wedesigna(1-e-1/4)-approximationalgorithmandalsoanefficientheuristicalgorithmwhenconsideringadditionalconstraints(e.g.,pathlength).BasedonthecurrentInternettopology,wediscovera“3540-alliance”subset(accountingonly6.8%)of52,079ASes/IXPs,whichcanprovidehighQoSguaranteesfor99.29%E2Econnections.
SubscriptionCoveringforRelevance-basedFilteringinContent-BasedPublish/SubscribeSystems KaiwenZhang(TechnischeUniversitatMunchen),VinodMuthusamy(IBMResearch),MohammadSadoghi(PurdueUniversity),Hans-ArnoJacobsen(UniversityofToronto)
Large-scaleapplicationsrequireascalabledatadisseminationservicewithadvancedfilteringcapabilities.Weproposetheuseofacontent-basedpublish/subscribesystemwithsupportfortop-kfilteringinthecontextofsuchapplications.Wefocusontheproblemoftop-ksubscriptionfiltering,whereapublicationisdeliveredonlytothekhighestscoringsubscribers.Thenaiveapproachtoperformfilteringearlyatthepublisheredgeworksonlyifcompleteknowledgeofthesubscriptionsisavailable,whichisnotcompatiblewiththewell-establishedcoveringoptimizationinscalablecontent-basedpublish/subscribesystems.Weproposeanefficientrank-covertechniquetoreconciletop-ksubscriptionfilteringwithcovering.Weextendthecoveringmodeltosupporttop-kanddescribeanovelalgorithmforforwardingsubscriptionstopublisherswhilemaintainingcorrectness.Finally,wecompareoursolutionstoabaselinecoveringsystem.Inatypicalsetting,ouroptimizedsolutionisscalableandprovidesover81%ofthecoveringbenefit.
WorkflowOptimizationinPAW MaximFilatov(UNIGE),VerenaKantere(UniversityofGeneva)
![Page 96: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/96.jpg)
96
Manyindustrialapplications,fromdomainssuchastelecommunication,webandsales,requiretoperformcomplexanalyticsacrossseveraldataprocessingsystems.Theperformanceofsuchanalyticsisusuallyexpressedinworkflows,anditisataskthatisbothlabor-intensiveandtime-consuming.Atthesametime,withincreasingamountsofdatatobeanalysed,theoptimizationofanalyticsworkflowsbecomescrucialforsatisfyingbusinessobjectives.Thispaperfocusesonworkflowoptimizationwithrespecttotimeefficiency,overmultipleexecutionengines,suchasatraditionalDBMS,aMapReduceengine,andascriptingengine.Thisconfigurationisemergingasacommonparadigmusedtocombineanalysisofunstructuredandstructureddata.WeproposeanoveloptimizationtechniqueaspartofoursystemcalledPAW(PlatformforAnalyticsWorkflows).Thistechniquecreatesalternativeworkflowstructuresandtheirexecutionplansbasedonequivalentcombinationsandordersofoperators.Thetechniqueemploysanexhaustiveandaheuristicalgorithmtosearchefficientlythespaceofequivalentworkflowstructuresandselecttheonewiththeoptimalexecutionplan.Wepresentathoroughexperimentalstudyandweshowcasetheefficiencyoftheproposedoptimizationtechniqueinafullyfledgedmulti-enginesystem,appliedonthreereal-worldapplicationsandtheirdata,aswellasonasyntheticbenchmark.
AFirstLookatInformationEntropy-BasedDataPricing XijunLi(ShanghaiJiaoTongUniversity),JianguoYao(ShanghaiJiaoTongUniversity),XueLiu(McGillUniversity),HaibingGuan(ShanghaiJiaoTongUniversity)
Distributionofintangibleinformationgoodsisexperiencingtremendousgrowthinrecentyears,whichhasfacilitatedablossomingofinformationgoodseconomics.Asbigdatadevelops,therearemoreandmoreinformationgoodsmarketsfordatatrading.Inthecurrentofdatapricingpoliciesindatatrading,therearemanymetricstomeasurethevalueofdatagoods,suchasthedatagenerationdate,datavolume,anddataintegrity,etc.However,itisverychallengingtoidentifytheamountofdatainformationanditsdistribution,andthecorrespondingdatapricinghasrarelybeendiscussed.Inthispaper,weproposeanewdatapricingmetric,i.e.,thedatainformationentropy,whichhelpstomakeareasonablepriceinthedatatrading.Wefirstdemonstrateadatainformationmeasurementmethodbasedoninformationentropy,andthenproposeapricingfunctionbasedontheresultofdatainformationmeasurement.Tocomprehensivelyunderstandthenewdatapricingmetricandfacilitateitsapplicationindatatrading,weverifytherationalityofthedatainformationmeasurementmethodandgivethreeconcretepricingfunctions.Itisthefirsttimetolookattheinformationentropy-baseddatapricing,whichcaninspiretheresearchconcerningthepricingmechanismofdatagoods,furtherpromotingthedevelopmentofdataproductsbusiness.
RestrospectiveLightweightDistributedSnapshotsUsingLooselySynchronizedClocks AlekseyCharapko(SUNYBuffalo),AilidaniAilijiang(SUNYBuffalo),MuratDemirbas(SUNYBuffalo),SandeepKulkarni(MichiganStateUniversity)
Inordertotakeaconsistentsnapshotofadistributedsystem,itisnecessarytocollateandalignlocallogsfromeachnodetoconstructapairwiseconcurrentcut.ByleveragingNTPsynchronizedclocks,andaugmentingthemwithlogicalclockcausalityinformation,Retroscopeprovidesalightweightsolutionfortakingunplannedretrospectivesnapshotsofpastdistributedsystemstates.Insteadofstoringamultiversioncopyoftheentiresystemdata,thisisachievedefficientlybymaintainingaconfigurable-sizeslidingwindow-logateachnodetocapturerecentoperations.Inadditiontoretrospectivesnapshots,Retroscopealsoprovidesincrementalandrollingsnapshotsthatutilizeanexistingsnapshottoreducethecostofconstructinganewsnapshotinproximity.Thiscapabilityisusefulforperformingstepwisedebuggingandroot-causeanalysis,andsupportingdataintegritymonitoringandcheckpoint-recovery.WeimplementRetroscopefortheVoldemortdistributeddatastoreandevaluateitsperformanceundervaryingworkloads.
Power-AwarePopulationProtocols ChuanXu(LRI(CNRS/UPSud)),JannaBurman(LRI(CNRS/UPSud)),JoffroyBeauquier(LRI(CNRS/UPSud))
Inthispaper,weproposeaformalenergymodelwhichallowsananalyticalstudyofenergyconsumption,forthefirsttimeinthecontextofpopulationprotocols(PP).InPP,anonymousandboundedmemoryagentsmoveunpredictablyandcommunicateinpairs.Inordertoillustratethepowerandtheusefulnessoftheproposedenergymodel,wedevelopanewpower-awareprotocol(EB-TTFM)forthetaskofdatacollection.Theanalyticalresultsshowthat,intermsofenergyconsumption,EB-TTFMoutperformsaknowndatacollectionprotocolundercertainconditions.Finally,wepresentalowerboundconcerningenergyconsumptionofanypossibledatacollectionprotocolinPP,whichalsojustifiestheefficiencyofEB-TTFM.
MultiPub:LatencyandCost-AwareGlobal-ScaleCloudPublish/Subscribe JulienGascon-Samson(McGillUniversity),JörgKienzle(McGillUniversity),BettinaKemme(McGillUniversity)
Topic-basedpub/subisawidelyusedcommunicationmechanismindistributedsystemsfortargetedinformationdisseminationbetweenlooselycoupledentities.Toscaledynamicallydependingonthecurrentcommunicationdemands,pub/servicescanbeconvenientlydeployedinthecloud.Toprovidefastdissemination,theservicecanbedistributedacrossmultiplecloudregions.Thearchitecturaldesignandrun-timedeploymentofsuchamiddlewareistricky,though,asitcanhaveasignificanteffectoncommunicationlatencyandcloud-basedcost.Inthispaper,weproposeMultiPub,aflexiblepub/submiddlewareforlatency-constrained,world-widedistributedapplicationsthatdynamicallyreconfiguresthecommunicationlayertoensureapredefinedmaximumlatencyforpublicationdisseminationwhileminimizingcloud-basedcosts.Thisisachievedbyroutingpublicationseitherthroughasingleoracrossmultiplecloudregions.WedemonstratetheeffectivenessofMultiPubbypresentingasetofexperimentsthatreportontheachievedcommunicationlatencyandcostsavingscomparedtotraditionalapproaches,aswellasaperformanceevaluation.
ReachabilityinBinaryMultithreadedProgramsIsPolynomial AlexanderMalkis(TechnischeUniversitätMünchen),SteffenBorgwardt(UCDavis)
Automaticfindingofbugsinmultithreadedprogramsisanimportantbutinherentlydifficulttask,eveninthefinite-stateinterleaving-semanticscase.Thecomplexityofthistaskhasonlybeenpartiallyexploredsofar.Wemeasurequantitiessuchasthediameter,whichisthelongestfinitedistancerealizableinthetransitiongraphoftheprogram,thelocaldiameter,whichisthemaximumdistancefromanyprogramstatetoanythread-localstate,andthecomputationalcomplexityofbugfinding.Forthesubclassofso-calledbinarymultithreadedprograms,weprovenewbounds:allthesequantitiesaremajorizedbyapolynomialand,incertaincases,byalinear,logarithmic,orevenconstantfunction.Ourboundspresentapreparationsteptowardsthecorrespondingpolynomial-boundclaimsforgeneralprograms.Theseclaimscontrastsharplywiththecommonbeliefthatthemainobstacletoanalyzingconcurrentprogramsistheexponentialstateexplosioninthenumberofthreads.
AnEvent-LevelAbstractionforAchievingEfficiencyandFairnessinNetworkUpdate
![Page 97: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/97.jpg)
97
TingQu(NationalUniversityofDefenseTechnology),DekeGuo(NationalUniversityofDefenseTechnology),XiaominZhu(NationalUniversityofDefenseTechnology),JieWu(TempleUniversity),XiaoleiZhou(NationalUniversityofDefenseTechnology),ZhongLiu(NationalUniversityofDefenseTechnology)
Changesofnetworkstateareacommonsourceofinstabilityinnetworks.Anupdateeventtypicallyinvolvesmultipleflowsthatcompetefornetworkresourcesatthecostofreschedulingandmigratingsomeexistingflows.Priornetworkupdatingschemestacklesuchflowsindependently,ratherthanastheentityofanupdateevent.Theyonlyoptimizetheflow-levelmetricsfortheflowsinvolvedinanupdateevent.Inthispaper,wepresentanevent-levelabstractionofnetworkupdatewhichgroupsflowsofanupdateeventandschedulesthemtogethertominimizetheeventcompletiontime(ECT).Wethenstudytheschedulingproblemofmultipleupdateeventsforachievinghighschedulingefficiencyandpreservingfairness.Thedesignedleastmigrationtrafficfirst(LMTF)methodschedulesallupdateeventsintheFIFOorder,butavoidshead-of-lineblockingbyrandomlyfine-tuningthequeueorderofsomeevents.Itcanconsiderablyreducetheupdatecost,theaverage,andtailECTsofallupdateevents.Inaddition,wedesignageneralparallel-LMTF(P-LMTF)methodtoguaranteefairnessandfurtherimproveschedulingefficiencyamongupdateevents.ItimprovestheLMTFmethodbyopportunisticallyupdatingmultipleeventssimultaneously.ThecomprehensiveevaluationresultsindicatethattheaverageECTofourapproachisupto10×fasterthantheflow-levelschedulingmethodfornetworkupdateevents,anditstailECTisupto6×faster.OurP-LMTFmethodincurs75%reductionintheaverageECTcomparedwithFIFOwhenthenetworkutilizationexceeds70%,anditachievesa42%reductionintailECT.
ShortPaper2:CloudandDataCenterSystemsandNetworksDCM:DynamicConcurrencyManagementforScalingn-TierApplicationsinCloud HuiChen(LouisianaStateUniversity),QingyangWang(LouisianaStateuniversity),BalajiPalanisamy(UniversityofPittsburgh),PengchengXiong(Hortonworks)
Scalingwebapplicationssuchase-commerceincloudbyaddingorremovingserversinthesystemisanimportantpracticetohandleworkloadvariations,withthegoalofachievingbothhighqualityofservice(QoS)andhighresourceefficiency.Throughextensivescalingexperimentsofann-tierapplicationbenchmark(RUBBoS),wehaveobservedthatscalingonlyhardwareresourceswithoutappropriateadaptationofsoftresourceallocations(e.g.,threadorconnectionpoolsize)ofeachserverwouldcausesignificantperformancedegradationoftheoverallsystembyeitherunder-orover-utilizingthebottleneckresourceinthesystem.Wedevelopadynamicconcurrencymanagement(DCM)frameworkwhichintegratessoftresourceallocationsintothesystemscalingmanagement.DCMintroducesamodelwhichdeterminesanear-optimalconcurrencysettingtoeachtierofthesystembasedonacombinationofoperationalqueuinglawsandonlineanalysisoffine-grainedmeasurementdata.WeimplementDCMasatwo-levelactuatorwhichscalesbothhardwareandsoftresourcesinann-tiersystemontheflywithoutinterruptingtheruntimesystemperformance.OurexperimentalresultsdemonstratethatDCMcanachievesignificantlymorestableperformanceandhigherresourceefficiencycomparedtothestate-of-the-arthardware-onlyscalingsolutions(e.g.,AmazonEC2-AutoScale)underrealisticburstyworkloadtraces.
MorePeak,LessDifferentiation:TowardsAPricing-awareOnlineControlFrameworkforInter-DatacenterTransfers WenxinLi(DalianUniversityofTechnology),XiaoboZhou(TiajjinUniversity),KeqiuLi(DalianUniversityofTechnology),HengQi(DalianUniversityofTechnology),DekeGuo(NationalUniversityofDefenceTechnology)
Theemergingdeploymentofgeographicallydistributeddatacenters(DCs)incursasignificantamountofdatatransfersovertheInternet.SuchtransfersaretypicallychargedbyInternetServiceProviders(ISPs)withthewidelyadoptedq-thpercentilechargingmodel.Insuchchargingmodel,thetimeslotswithtop100-qpercentofdatatransmissiondonotaffectthetotaltransmissioncost,andcanbeviewedasfree.Thisbringstheopportunitytooptimizetheschedulingofinter-DCtransferstominimizetheentiretransmissioncost.However,verylittleworkhasbeendonetoexploitthosefreetimeslotsforschedulinginter-DCtransfers.Thecruxisthatexistingworkeitherlacksamechanismtoaccumulatetraffictofreetimeslots,orinevitablyreliesonpriorknowledgeoftrafficarrivalpatterns.Inthispaper,weattempttoexploitthosefreetimeslotsbyleveragingdiversetime-sensitivitiesamonginter-DCtransfers,soastoreduceorevenminimizethetransmissioncost.Specifically,weadvocatethatasimpleprincipleshouldbefollowed:moretrafficpeaksshouldbescheduledinfreetimeslots,whilelesstrafficdifferentiationshouldbemaintainedamongtheremainingtimeslots.Tothisend,wetakeadvantageoftheLyapunovoptimizationtechniquestodesignapricing-awarecontrolframework.Thisframeworkefficientlymakesonlinedecisionsforinter-DCtransferswithoutrequiringapriorknowledgeoftrafficarrivals.Toverifyourproposedframework,weconductsmall-scaletestbedimplementation.Theresultsshowthatourframeworkcanrealisticallyreducethetransmissioncostbyupto19.38%.
RobustMulti-TenantServerConsolidationintheCloudforDataAnalyticsWorkloads JosephMate(UniversityofWaterloo),KhuzaimaDaudjee(UniversityofWaterloo),ShahinKamali(MITCSAIL)
Serverconsolidationistheallocationorhostingoftenantsonaminimumnumberofcloudservermachines.Givenasequenceofdataanalyticstenantloadsdefinedbytheamountofresourcesthatthetenantsrequireandaservice-levelagreement(SLA)betweenthecustomerandthecloudserviceprovider,significantresourcecostsavingscanbeachievedbyconsolidatingmultipletenantsonservermachines.Sinceservermachinescanfailcausingtheirtenantstobecomeunavailable,serviceproviderscanplacereplicasofeachtenantondifferentserversandreservecapacitytoensurethattenantfailoverwillnotresultinoverloadonanyremainingserver.WeproposetheCUBEFITalgorithmformulti-tenantserverconsolidationthatsavesresourcecostsbyutilizingfewerserversthanexistingapproachesfordataanalyticsworkloads.Unlikeexistingconsolidationalgorithms,CUBEFITcantoleratemultipleserverfailureswhileensuringthatnoserverbecomesoverloaded.WeprovideextensivetheoreticalanalysisandexperimentalevaluationofCUBEFIT.Weshowthatcomparedtoexistingalgorithms,theaveragecaseandworstcasebehaviorofCUBEFITissuperiorandthatitproducesnear-optimaltenantallocationwhenthenumberoftenantsislarge.Throughevaluationanddeploymentonaclusterofupto73machinesaswellasthroughsimulationstudies,weexperimentallydemonstratetheefficacyofCUBEFIT.
Flow-AwareAdaptivePacingtoMitigateTCPIncastinDatacenterNetworks ShaojunZou(CentralSouthUniversity),JiaweiHuang(CentralSouthUniversity),YutaoZhou(CentralSouthUniversity),JianxinWang(CentralSouthUniversity),TianHe(UniversityofMinnesota)
Indatacenternetworks,manynetwork-intensiveapplicationsleveragelargefan-inandmany-to-onecommunicationtoachievehighperformance.However,thespecialtrafficpatterns,suchasmicro-burstandhighconcurrency,easilycauseTCPIncastproblemandseriouslydegradetheapplicationperformance.ToaddresstheTCPIncastproblem,wefirstrevealtheoreticallyandempiricallythatalleviatingpacketburstinessismuchmoreeffectiveinreducingtheIncastprobabilitythancontrollingthecongestionwindow.Inspiredbythefindingsandinsightsfromourexperimentalobservations,wefurtherproposeageneralsupportingschemeAdaptivePacing(AP),whichdynamicallyadjustsburstinessaccordingtotheflowconcurrencywithoutanychangeonswitch.AnotherfeatureofAPisitsbroad
![Page 98: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/98.jpg)
98
applicability.WeintegrateAPtransparentlyintodifferentTCPprotocols(i.e.,DCTCP,L2DCTandD2TCP).Throughaseriesoflarge-scaleNS2simulations,weshowthatAPsignificantlyreducestheIncastprobabilityacrossdifferentTCPprotocolsandthenetworkgoodputcanbeincreasedconsistentlybyonaverage7xunderseverecongestion.
Real-TimePowerCyclinginVideoonDemandDataCentresusingOnlineBayesianPrediction VicentSanzMarco(LancasterUniversity),ZhengWang(LancasterUniversity),BarryPorter(LancasterUniversity)
Energyusageindatacentrescontinuestobeamajorandgrowingconcernasanincreasingnumberofeverydayservicesdependonthesefacilities.Researchinthisareahasexaminedtopicsincludingpowersmoothingusingbatteriesanddeeplearningtocontrolcoolingsystems,inadditiontooptimisationtechniquesforthesoftwarerunninginsidedatacentres.Wepresentanovelreal-timepower-cyclingarchitecture,supportedbyamediadistributionapproachandonlinepredictionmodel,toautomaticallydeterminewhenserversareneededbasedondemand.Wedemonstratewithexperimentalevaluationthatthisapproachcansaveupto31%ofserverenergyinacluster.Ourevaluationisconductedontypicalrackmountserversinadatacentretestbedandusesarecentreal-worldworkloadtracefromtheBBCiPlayer,anextremelypopularvideoondemandserviceintheUK.
ADistributedAccessControlSystemforCloudFederations ShorouqAlansari(UniversityofSouthampton),FedericaPaci(UniversityofSouthampton),VladimiroSassone(UniversityofSouthampton)
Cloudfederationsareanewcollaborationparadigmwhereorganizationssharedataacrosstheirprivatecloudinfrastructures.However,theadoptionofcloudfederationsishinderedbyfederatedorganizations’concernsonpotentialrisksofdataleakageanddatamisuse.Forcloudfederationstobeviable,federatedorganizations’privacyconcernsshouldbealleviatedbyprovidingmechanismsthatalloworganizationstocontrolwhichusersfromotherfederatedorganizationscanaccesswhichdata.Weproposeanovelidentityandaccessmanagementsystemforcloudfederations.Thesystemallowsfederatedorganizationstoenforceattribute-basedaccesscontrolpoliciesontheirdatainaprivacy-preservingfashion.Usersaregrantedaccesstofederateddatawhentheiridentityattributesmatchthepolicies,butwithoutrevealingtheirattributestothefederatedorganizationowningdata.ThesystemalsoguaranteestheintegrityofthepolicyevaluationprocessbyusingblockchaintechnologyandIntelSGXtrustedhardware.Itusesblockchaintoensurethatusersidentityattributesandaccesscontrolpoliciescannotbemodifiedbyamalicioususer,whileIntelSGXprotectstheintegrityandconfidentialityofthepolicyenforcementprocess.Wepresenttheaccesscontrolprotocol,thesystemarchitectureanddiscussfutureextensions.
Voyager:CompleteContainerStateMigration ShripadNadgowda(IBMTJWatsonResearchCenter),SahilSuneja(IBMTJWatsonResearchCenter),NiltonBila(IBMTJWatsonResearchCenter),CanturkIsci(IBMTJWatsonResearchCenter)
Duetothesmallmemoryfootprintandfaststartuptimesofferredbycontainervirtualization,madeevermorepopularbytheDockerplatform,containersareseeingrapidadoptionasafoundationalcapabilitytobuildPaaSandSaaSclouds.Forsuchcontainerclouds,whicharefundamentallydifferentfromVMclouds,variouscloudmanagementservicesneedtoberevisited.Inthispaper,wepresentourVoyager-just-in-timelivecontainermigrationservice,designedinaccordancewiththeOpenContainerInitiative(OCI)principles.Voyagerisanovelfilesystem-agnosticandvendor-agnosticmigrationservicethatprovidesconsistentfull-systemmigration.VoyagercombinesCRIU-basedmemorymigrationtogetherwiththedatafederationcapabilitiesofunionmountstominimizemigrationdowntime.Withaunionviewofdatabetweenthesourceandtargethosts,Voyagercontainerscanresumeoperationinstantlyonthetargethost,whileperformingdiskstatetransferlazilyinthebackground.
Keddah:CapturingHadoopNetworkBehaviour JieDeng(QueenMaryUniversityLondon),GarethTyson(QueenMary),FélixCuadrado(QueenMaryUniversityofLondon),SteveUhlig(QueenMaryUniversityofLondon)
Asadistributedsystem,Hadoopheavilyreliesonthenetworktocompletedataprocessingjobs.WhileHadooptrafficisperceivedtobecriticalforjobexecutionperformance,theactualbehaviourofHadoopnetworktrafficisstillpoorlyunderstood.ThislackofunderstandinggreatlycomplicatesresearchrelyingonHadoopworkloads.Inthispaper,weexploreHadooptrafficthroughexperimentation.WeanalysethegeneratedtrafficofmultipletypesofMapReducejobs,withvaryinginputsizes,andclusterconfigurationparameters.Asaresult,wepresentKeddah,atoolchainforcapturing,modellingandreproducingHadooptraffic,forusewithnetworksimulators.KeddahcanbeusedtocreateempiricalHadooptrafficmodels,enablingreproducibleHadoopresearchinmorerealisticscenarios.
AScalableandDistributedApproachforNFVServiceChainCostMinimization ZijunZhang(UniversityofCalgary),ZongpengLi(UniversityofCalgary),ChuanWu(UniversityofHongKong),ChuanheHuang(WuhanUniversity)
Networkfunctionvirtualization(NFV)representsthelatesttechnologyadvancementinnetworkserviceprovisioning.Traditionalhardwaremiddleboxesarereplacedbysoftwareprogramsrunningonindustrystandardserversandvirtualmachines,forserviceagility,flexibility,andcostreduction.NFVusersareprovisionedwithservicechainscomposedofvirtualnetworkfunctions(VNFs).AfundamentalprobleminNFVservicechainprovisioningistosatisfyuserdemandswithminimumsystem-widecost.Wejointlyconsidertwotypesofcostinthiswork:nodalresourcecostandlinkdelaycost,andformulatetheservicechainprovisioningproblemusingnonlinearoptimization.Throughthemethodofauxiliaryvariables,wetransformtheoptimizationproblemintoitsseparableform,andthenapplythealternatingdirectionmethodofmultipliers(ADMM)todesignscalableandfullydistributedsolutions.Throughsimulationstudies,weverifytheconvergenceandefficacyofourdistributedalgorithmdesign.
ElasticPaxos:ADynamicAtomicMulticastProtocol SamuelBenz(UniversitàdellaSvizzeraitaliana),FernandoPedone(UniversitàdellaSvizzeraitaliana)
Replicationisacommontechniqueusedtodesignreliabledistributedsystemsbymaskingdefectivecomponents.TocopewiththerequirementsofmodernInternetapplications,replicationprotocolsmustallowforthroughputscalabilityanddynamicreconfiguration,thatis,on-demandreplacementorprovisioningofsystemresources.ThispaperdescribesElasticPaxos,anewdynamicatomicmulticastprotocolthatfulfillstheserequirements.ElasticPaxosallowstodynamicallyaddandremoveresourcestoanonlinepartiallyreplicatedstatemachine.WeimplementedElasticPaxosandevaluateditsperformanceinOpenStack,acloudenvironment.Wedemonstrateitspracticalitytodynamicallyscaleupanddownapartiallyreplicateddatastorewithitandtoreconfigureadistributedsystem.
BoostingTheBenefitsOfHybridSDN WenWang(McGillUniversity),WenboHe(McMasterUniversity),JinshuSu(NationalUniversityofDefenseTechnology)
![Page 99: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/99.jpg)
99
Theupgradeofalegacynetworktoafullsoftware-definednetworking(SDN)deploymentisusuallyanincrementalprocess,duringwhichSDNswitchesandlegacyswitchescoexistinthehybridnetwork.However,withinappropriatedeploymentofSDNswitchesanddesignofhybridcontrol,theadvantagesofSDNcontrolcouldnotexert,anditevenresultsinperformancedegradationorinconsistency(e.g.,loops,black-holes).Therefore,thehybridSDNrequiresconsiderablecoordinationofthecentralizedcontrolanddistributedrouting.Inthispaper,weproposeasolutiontohandletheheterogeneitycausedbydistinctforwardingcharacteristicsofSDNandlegacyswitches,thereforeboostingthebenefitsofhybridSDN.WeplanSDNplacementtoenhancetheSDNcontrollabilityoverthehybridnetwork,andconducttrafficengineeringconsideringboththeforwardingcharacteristicsofSDNandlegacyswitches.TheexperimentswithvarioustopologiesshowthattheSDNplacementplanningandhybridforwardingyieldbetternetworkperformanceespeciallyintheearly70%SDNdeployment.
AdoptingSDNSwitchBuffer:BenefitsAnalysisandMechanismDesign FuliangLi(NortheasternUniversity),JiannongCao(TheHongKongPolytechnicUniversity),XingweiWang(NortheasternUniversity),YinchuSun(NortheasternUniversity),TianPan(BeijingUniversityofPostsandTelecommunication),XuefengLiu(TheHongKongPolytechnicUniversity)
OnecriticalissueinSDNistoreducethecommunicationoverheadbetweentheswitchesandthecontroller.Suchoverheadismainlycausedbyhandlingmiss-matchpackets,becauseforeachmiss-matchpacket,aswitchwillsendarequesttothecontrolleraskingforforwardingrule.Existingapproachestoaddressthisproblemgenerallyneedtodeployintermediateproxyorauthorityswitchestoholdrulecopies,soastoreducethenumberofrequestssenttothecontroller.Inthispaper,wearguethatusingtheintrinsicbufferinaSDNswitchcanalsogreatlyreducethecommunicationoverheadwithoutusingadditionaldevices.Ifaswitchbufferseachmiss-matchpacket,onlyafewheaderfieldsinsteadoftheentirepacketarerequiredtobesenttothecontroller.Experimentresultsshowthatthiscanreduce78.7%controltrafficand37%controlleroverheadatthecostofincreasingonly5.6%switchoverheadonaverage.Iftheproposedflow-granularitybuffermechanismisadopted,onlyonerequestmessageneedstobesenttothecontrollerforanewflowwithmanyarrivalpackets.Thusthecontroltrafficandcontrolleroverheadcanbefurtherreducedby64%and35.7%respectivelyonaveragewithoutincreasingtheswitchoverhead.
ShortPaper3:InternetofThings,SmartCities,andCyber-PhysicalSystemsIOTSENTINEL:AutomatedDevice-TypeIdentificationforSecurityEnforcementinIoT MarkusMiettinen(TechnischeUniversitatDarmstadt),SamuelMarchal(AaltoUniversity),IbbadHafeez(UniversityofHelsinki),N.Asokan(AaltoUniversity),Ahmad-RezaSadeghi(TechnischeUniversitatDarmstadt),SasuTarkoma(UniversityofHelsinki)
WiththerapidgrowthoftheInternet-of-Things(IoT),concernsaboutthesecurityofIoTdeviceshavebecomeprominent.SeveralvendorsareproducingIP-connecteddevicesforhomeandsmallofficenetworksthatoftensufferfromflawedsecuritydesignsandimplementations.Theyalsotendtolackmechanismsforfirmwareupdatesorpatchesthatcanhelpeliminatesecurityvulnerabilities.Securingnetworkswherethepresenceofsuchvulnerabledevicesisgiven,requiresabrownfieldapproach:applyingnecessaryprotectionmeasureswithinthenetworksothatpotentiallyvulnerabledevicescancoexistwithoutendangeringthesecurityofotherdevicesinthesamenetwork.Inthispaper,wepresentIoTSentinel,asystemcapableofautomaticallyidentifyingthetypesofdevicesbeingconnectedtoanIoTnetworkandenablingenforcementofrulesforconstrainingthecommunicationsofvulnerabledevicessoastominimizedamageresultingfromtheircompromise.WeshowthatIoTSentineliseffectiveinidentifyingdevicetypesandhasminimalperformanceoverhead.
EfficientZ-orderEncodingBasedMulti-modelDataCompressioninWSNs XiaofeiCao(MissouriUniversityofScienceandTechnology),SanjayMadria(MissouriUniversityofScienceandTechnology),TakahiroHara(OsakaUniversity)
Wirelesssensornetworkshavesignificantlimitationsinavailablebandwidthandenergy.Thelimitedbandwidthinsensornetworkscancausehighermessagedeliverylatencyinapplicationssuchasmonitoringpoisonousgasleak.Insuchapplications,therearemulti-modalsensorswhosevaluessuchastemperature,gasconcentration,locationandCO2levelneedtobetransmittedtogetherforfasterdetectionandtimelyassessmentofgasleak.Inthispaper,weproposenovelZ-orderbaseddatacompressionschemes(Z-compression)toreduceenergyandsavebandwidthwithoutincreasingthemessagedeliverylatency.InsteadofusingthepopularHuffmantreestylebasedencoding,ZcompressionusesZ-orderencodingtomapthemultidimensionalsensingdataintoone-dimensionalbinarystreamtransmittedusingasinglepacket.Ourexperimentalevaluationsusingreal-worlddatasetsshowthatZ-compressionhasamuchbettercompressionratio,energysaving,streamingratethanknownschemeslikeLEC(andadaptiveLEC),FELACSandTinyPackformulti-modalsensordata.
PTrack:EnhancingtheApplicabilityofPedestrianTrackingwithWearables YonghangJiang(CityUniversityofHongKong),ZhenjiangLi(CityUniversityofHongKong),JianpingWang(CityUniversityofHongKong)
Theabilitytoaccuratelytrackpedestriansisvaluableforvariantapplicationdesigns.Althoughpedestriantrackinghasbeeninvestigatedexcessivelyandownedawell-suitedsensingplatform,theproposedsolutionsarefarfrombeingmatureyet.Pedestriantrackingcontainsstepcountingandstrideestimationtwocomponents.Stepcountingalreadyhascommercialproducts,buttheperformanceisstillunreliableandlesstrustworthyinpractice.Strideestimationevenstaysintheresearchstagewithoutreadysolutionsreleasedonthemarket.Suchanon-negligiblegapbetweenlong-termresearchinvestigationandtechnique'sactualusageexistsduetoaseriesofcrucialapplicabilityissuesunsolved,includingdesignvulnerabilitytointerferingactivities,extractingpurelybody'smovementfromadditivesensorsignals,andparametertrainingwithoutuser'sintervention.Inthispaper,wedeeplyanalyzehuman'sgaitcyclesandobtaininspiringobservationstoaddresstheseissues.Weincorporateourtechniquesintoexistingpedestriantrackingdesignsandimplementaprototype,PTrack,onLGsmartwatch.WefindPTrackeffectivelyenhancesthesystemapplicabilityandachievespromisingperformanceunderverypracticalsettings.
SourceLocationPrivacy-AwareDataAggregationSchedulingforWirelessSensorNetworks JackKirton(UniversityofWarwick),MatthewBradbury(TheUniversityofWarwick),ArshadJhumka(UniversityofWarwick)
Sourcelocationprivacy(SLP)isanimportantpropertyfortheclassofassetmonitoringproblemsinwirelesssensornetworks(WSNs).SLPaimstopreventanattackerfromfindingavaluableassetwhenaWSNnodeisbroadcastinginformationduetothedetectionoftheasset.MostSLPtechniquesfocusattheroutinglevel,withtypicallyhighmessageoverhead.TheobjectiveofthispaperistoinvestigatethenovelproblemofdevelopingaTDMAMACschedulethatcanprovideSLP.Wemakeanumberofimportantcontributions:(i)wedevelopanovelformalisationofaclassofeavesdroppingattackersandprovidenovelformalisationsofSLP-awaredataaggregationschedules(DAS),(ii)wepresentadecisionproceduretoverifywhetheraDASscheduleisSLP-aware,thatreturnsacounterexampleifthescheduleisnot,similartomodelchecking,and(iii)wedevelopa3-stagedistributedalgorithmthattransformsaninitialDASalgorithmintoacorrespondingSLP-awarescheduleagainstaspecificclassofeavesdroppers.OursimulationresultsshowthattheresultingSLP-awareDASprotocolreducesthecaptureratioby50%attheexpenseofnegligablemessageoverhead.
![Page 100: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/100.jpg)
100
VelocityOptimizationofPureElectricVehicleswithTrafficDynamicsConsideration LiuwangKang(UniversityofVirginia),HaiyingShen(UniversityofVirginia),AnkurSarker(UniversityofVirginia)
AsElectricVehicles(EVs)becomeincreasinglypopular,theirbattery-relatedproblems(e.g.,shortdrivingrangeandheavybatteryweight)mustberesolvedassoonaspossible.VelocityoptimizationofEVstominimizeenergyconsumptionindrivingisaneffectivealternativetohandletheseproblems.However,previousvelocityoptimizationmethodsassumethatvehicleswillpassthroughtrafficlightsimmediatelyatgreentrafficsignals.Actually,avehiclemaystillexperienceadelaytopassagreentrafficlightduetoavehiclewaitingqueueinfrontofthetrafficlight.Inthispaper,forthefirsttime,weproposeavelocityoptimizationsystemwhichenablesEVstoimmediatelypassgreentrafficlightswithoutdelay.Wecollectedrealdrivingdataona4.0kmlongroadsectionofUS-25highwaytoconductextensivetrace-drivensimulationstudies.TheexperimentalresultsfromMatlabandSimulationforUrbanMObility(SUMO)trafficsimulatorshowthatourvelocityoptimizationsystemreducesenergyconsumptionbyupto17.5%comparedwithrealdrivingpatternswithoutincreasingtriptime.
PIANO:Proximity-basedUserAuthenticationonVoice-PoweredInternet-of-ThingsDevices NeilZhenqiangGong(IowaStateUniversity),AltayOzen(IowaStateUniversity),YuWu(UCDavis),XiaoyuCao(IowaStateUniversity),RichardShin(UCBerkeley),DawnSong(UCBerkeley),HongxiaJin(SamsungResearchAmerica),XuanBao(GoogleInc.)
VoiceisenvisionedtobeapopularwayforhumanstointeractwithInternet-of-Things(IoT)devices.Weproposeaproximity-baseduserauthenticationmethod(calledPIANO)foraccesscontrolonsuchvoice-poweredIoTdevices.PIANOleveragesthebuilt-inspeaker,microphone,andBluetooththatvoice-poweredIoTdevicesoftenalreadyhave.Specifically,weassumethatausercarriesapersonalvoice-powereddevice(e.g.,smartphone,smartwatch,orsmartglass),whichservesastheuser’sidentity.Whenanothervoice-poweredIoTdeviceoftheuserrequiresauthentication,PIANOestimatesthedistancebetweenthetwodevicesbyplayinganddetectingcertainacousticsignals;PIANOgrantsaccessiftheestimateddistanceisnolargerthanauser-selectedthreshold.Weimplementedaproof-of-conceptprototypeofPIANO.Throughtheoreticalandempiricalevaluations,wefindthatPIANOissecure,reliable,personalizable,andefficient.
CategoryInformationCollectioninRFIDSystems JiaLiu(NanjingUniversity),ShigangChen(UniversityofFlorida),BinXiao(TheHongKongPolytechnicUniversity),YanyanWang(NanjingUniversity),LijunChen(NanjingUniversity)
InRFID-enabledapplications,whenatagisputintouseandassociatedwithaspecificobject,thecategory-relatedinformation(e.g.,thebrandsofclothes)aboutthisobjectmightbepreloadedintothetag’smemoryasrequired.Sincesuchinformationreflectsthecategoryattributes,alltagsinthesamecategorycarrytheidenticalcategoryinformation.Tocollectthisinformation,wedonotneedtorepeatedlyinterrogateeachtag;onetag’sresponseinacategoryissufficient.Inthispaper,weinvestigatethenewproblemofcategoryinformationcollectioninamulti-categoryRFIDsystem,whichisreferredtoasinformationsampling.Weproposeanefficienttwo-phasesamplingprotocol(TPS).Byquicklyzoomingintoacategoryandisolatingatagfromthiscategory,TPSisabletosampleacategorybybroadcastingonly7.5-bitpollingvector(veryefficientwhencomparedtothe96-bittagID).Wetheoreticallyanalyzetheprotocolperformanceanddiscusstheoptimalparametersettingsthatminimizetheoverallexecutiontime.ExtensivesimulationsshowthatTPSoutperformsthebenchmark,greatlyimprovingthesamplingperformance.
ScalableRole-basedDataDisclosureControlfortheInternetofThings AliYavari(RMITUniversity),ArezouSoltaniPanah(RMITUniversity),DimitriosGeorgakopoulos(SwinburneUniversityofTechnology),PremPrakashJayaraman(SwinburneUniversityofTechnology),RonvanSchyndel(RMITUniversity)
TheInternetofThings(IoT)isthelatestInternetevolutionthatinterconnectsbillionsofdevices,suchascameras,sensors,RFIDs,smartphones,wearabledevices,ODBIIdongles,etc.FederationsofsuchIoTdevices(or\textit{things})providestheinformationneededtosolvemanyimportantproblemsthathavebeentoodifficulttoharnessbefore.Despitethesegreatbenefits,privacyinIoTremainsagreatconcern,inparticularwhenthenumberofthingsincreases.Thispressestheneedforthedevelopmentofhighlyscalableandcomputationallyefficientmechanismstopreventunauthorisedaccessanddisclosureofsensitiveinformationgeneratedbythings.Inthispaper,weaddressthisneedbyproposingalightweight,yethighlyscalable,dataobfuscationtechnique.Forthispurpose,adigitalwatermarkingtechniqueisusedtocontrolperturbationofsensitivedatathatenableslegitimateuserstode-obfuscateperturbeddata.Toenhancethescalabilityofoursolution,wealsointroduceacontextualisationservicethatachievereal-timeaggregationandfilteringofIoTdataforlargenumberofdesignatedusers.We,then,assesstheeffectivenessoftheproposedtechniquebyconsideringahealth-carescenariothatinvolvesdatastreamedfromvariouswearableandstationarysensorscapturinghealthdata,suchasheart-rateandbloodpressure.Ananalysisoftheexperimentalresultsthatillustratetheunconstrainedscalabilityofourtechniqueconcludesthepaper.
Multi-representationbasedDataProcessingArchitectureforIoTApplications VaibhavArora(UniversityofCalifornia,SantaBarbara),FaisalNawab(UniversityofCalifornia,SantaBarbara),DivyakantAgrawal(UniversityofCalifornia,SantaBarbara),AmrElAbbadi(UniversityofCalifornia,SantaBarbara)
InternetofThings(IoT)applicationslikesmartcars,smartcitiesandwearablesarebecomingwidespreadandarethefutureoftheInternet.OneofthemajorchallengesforIoTapplicationsisefficientlyprocessing,storingandanalyzingthecontinuousstreamofincomingdatafromalargenumberofconnectedsensors.Weproposeamulti-representationbaseddataprocessingarchitectureforIoTapplications.Thedataisstoredinmultiplerepresentations,likerows,columns,graphswhichprovidessupportfordiverseapplicationdemands.Aunifyingupdatemechanismbasedondeterministicschedulingisusedtoupdatethedatarepresentations,whichcompletelyremovestheneedfordatatransferpipelineslikeETL(Extract,TransformandLoad).Thecombinationofmultiplerepresentations,andthedeterministicupdatemechanism,providestheabilitytosupportreal-timeanalyticsandcaterstoIoTapplicationsbyminimizingthelatencyofoperationslikecomputingpre-definedaggregates.
LongTermSensingviaBatteryHealthAdaptation GregJackson(ImperialCollegeLondon),ZhijinQin(ImperialCollegeLondon),JulieAMcCann(ImperialCollegeLondon)
EnergyNeutralOperation(ENO)hascreatedtheabilitytocontinuouslyoperatewirelesssensornetworksinareassuchasenvironmentalmonitoring,hazarddetectionandindustrialIoTapplications.CurrentENOapproachesutilisetechniquessuchassampleratecontrol,adaptivedutycyclinganddatareductionmethodstobalanceenergygeneration,storageandconsumption.However,thestateoftheartapproachesmakesastrongandunrealisticassumptionthatbatterycapacityisfixedthroughoutthedeploymenttimeofanapplication.ThisresultsinscenarioswhereENOsystemsoverallocatesensingtasks,thereforeasbatterycapacity
![Page 101: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/101.jpg)
101
degradesitcausesthesystemtonolongerbeenergyneutralandthenfailunexpectedly.Inthispaper,weformulatetheproblemtomaximisethequality-of-serviceintermsofdutycycleandthebatterycapacitytoextendthedeploymentlifetimeofasensingapplication.Inaddition,wedevelopalightweightalgorithmtosolvetheformulatedproblem.Moreover,weevaluatetheproposedmethodusingrealsensorenergyconsumptiondatacapturedfrommicro-climatesensorsdeployedinQueenElizabethOlympicPark,London.Resultsshowthata307%extensionofdeploymentlifetimecanbeachievedwhencomparedtoatraditionalENOsolutionwithoutareductioninthedutycycleofthesensor.
DetectingTimeSynchronizationAttacksinCyber-PhysicalSystemswithMachineLearningTechniques JingxuanWang(TheUniversityofHongKong),WentingTu(ShanghaiUniversityofFinanceandEconomics),LucasC.K.Hui(TheUniversityofHongKong),SiuMingYiu(TheUniversityofHongKong),EricKeWang(HarbinInstituteofTechnologyShenzhenGraduateSchool)
Recently,researchersfoundanewtypeofattacks,calledtimesynchronizationattack(TSattack),incyber-physicalsystems.Insteadofmodifyingthemeasurementsfromthesystem,thisattackonlychangesthetimestampsofthemeasurements.Studiesshowthattheseattacksarerealisticandpractical.However,existingdetectiontechniques,e.g.baddatadetection(BDD)andmachinelearningmethods,maynotbeabletocatchtheseattacks.Inthispaper,wedevelopa“firstdifferenceaware”machinelearning(FDML)classifiertodetectthisattack.Thekeyconceptbehindourclassifieristousethefeatureof“firstdifference”,borrowedfromeconomicsandstatistics.SimulationsonIEEE14-bussystemwithrealdatafromNYISOhaveshownthatourFDMLclassifiercaneffectivelydetectbothTSattacksandothercyberattacks.
Speed-basedLocationTrackinginUsage-basedAutomotiveInsurance LuZhou(ShanghaiJiaoTongUniversity),QingrongChen(ShanghaiJiaoTongUniversity),ZutianLuo(ShanghaiJiaoTongUniversity),HaojinZhu(ShanghaiJiaoTongUniversity),CailianChen(ShanghaiJiaoTongUniversity)
Usage-basedInsurance(UBI)isregardedasapromisingwaytooffermoreaccurateinsurancepremiumbyprofilingdrivingbehaviors.Comparedwithtraditionalinsurancewhichconsidersdrivers’historyofaccidents,trafficviolationsandetc,UBIfocusesondrivingdataandcangiveamorereasonableinsurancepremiumbasedonthecurrentdrivingbehaviors.Insurersusesensorsinsmartphoneorvehicletocollectdrivingdata(e.g.mileage,speed,harkbraking)andcomputeariskscorebasedonthesedatatorecalculateinsurancepremium.Manyinsuranceprograms,whichareadvertisedasbeingprivacy-preserving,donotdirectlyusetheGPS-basedtracking,butitisnotenoughtoprotectdriver’slocationprivacy.Inrealworld,manyenvironmentfactorssuchasreal-timetrafficandtrafficregulationscaninfluencedrivingspeed.Thesefactorsprovidetheside-channelinformationaboutthedrivingroute,whichcanbeexploitedtoinferthevehicle’strace.Basedontheobservation,weproposeanovelspeedbasedtrajectoryinferencealgorithmwhichcantrackdriversonlywiththespeeddataandoriginallocation.WeimplementtheattackonapublicdatasetinNewJersey.Theevaluationresultsshowthattheattackercanrecovertheroutewithahighsuccessfulrate.
ShortPaper4:Mobile,Wireless,Edge,andCrowdComputingOnefficientoffloadingcontrolincloudradioaccessnetworkwithmobileedgecomputing TongLi(TsinghuaUniversity),ChathuraSarathchandraMagurawalage(UniversityofEssex),KezhiWang(UniversityofEssex),KeXu(TsinghuaUniversity),KunYang(UniversityofEssex),HaiyangWang(UniversityofMinnesotaatDuluth)
Cloudradioaccessnetwork(C-RAN)andmobileedgecomputing(MEC)haveemergedaspromisingcandidatesforthenextgenerationaccessnetworktechniques.Unfortunately,althoughMECtriestoutilizethehighlydistributedcomputingresourcesincloseproximitytouserequipmentsequipments(UE),C-RANsuggeststocentralizethebasebandprocessingunits(BBU)deployedinradioaccessnetworks.Tobetterunderstandandaddresssuchaconflict,thispapercloselyinvestigatestheMECtaskoffloadingcontrolinC-RANenvironments.Inparticular,wefocusonperspectiveofmatchingproblem.OurmodelsmartlycapturestheuniquefeaturesinbothMECandC-RANwithrespecttocommunicationandcomputationefficiencyconstraints.Wedividethecross-layeroptimizationintothefollowingthreestages:(1)matchingbetweenremoteradioheads(RRH)andUEs,(2)matchingbetweenBBUsandUEs,and(3)matchingbetweenmobileclones(MC)andUEs.ByapplyingtheGale-ShapleyMatchingTheoryintheduplexmatchingframework,weproposeamulti-stageheuristictominimizetherefusalrateforuser'staskoffloadingrequests.Trace-basedsimulationconfirmsthatoursolutioncansuccessfullyachievenear-optimalperformanceinsuchahybriddeployment.
LocationPrivacyinMobileEdgeClouds TingHe(PennsylvaniaStateUniversity),ErtugrulCiftcioglu(ArmyResearchLaboratory),ShiqiangWang(IBM),KevinChan(ArmyResearchLaboratory)
Inthispaper,weconsideruserlocationprivacyinmobileedgeclouds(MECs).MECsaresmallcloudsdeployedatthenetworkedgetooffercloudservicesclosetomobileusers,andmanysolutionshavebeenproposedtomaximizeservicelocalitybymigratingservicestofollowtheirusers.Co-locationofauserandhisservice,however,impliesthatacybereavesdropperobservingservicemigrationsbetweenMECscanlocalizetheuseruptooneMECcoveragearea,whichcanbefairlysmall(e.g.,afemtocell).Weconsiderusingchaffservicestodefendagainstsuchaneavesdropper,withfocusonstrategiestocontrolthechaffs.Assumingtheeavesdropperperformsmaximumlikelihood(ML)detection,weconsiderbothheuristicstrategiesthatmimictheuser'smobilityandoptimizedstrategiesdesignedtominimizethedetectionortrackingaccuracy.Weshowthatasinglechaffcontrolledbytheoptimalstrategycandrivetheeavesdropper'strackingaccuracytozerowhentheuser'smobilityissufficientlyrandom.Theefficacyofoursolutionsisverifiedthroughextensivesimulations.
ApproximationDesignforCooperativeRelayDeploymentinWirelessNetworks HaotianWang(ShanghaiJiaoTongUniversity),ShileiTian(ShanghaiJiaoTongUniversity),XiaofengGao(ShanghaiJiaoTongUniversity),LidongWu(UniversityofTexasatTyler),GuihaiChen(ShanghaiJiaoTongUniversity)
Inthispaper,weaimtomaximizeusers’satisfactionbydeployinglimitednumberofrelaysinatargetregiontoformawirelessrelaynetwork,anddefinetheDeploymentofCooperativeRelay(DoCR)problem,whichisprovedtobeNP-complete.WefirstproposeanO(δlogn)approximationalgorithmthatutilizesthealgorithmsforbudgetweightedSteinertreeproblemwithnovelpositionweightingassignment.WefurtherproposeaheuristicmethodtosolvetheDoCRproblemreleasingpotentiallocationconstraint.Ourextensiveexperimentsindicatethatthealgorithmsweproposecansignificantlyimprovethetotalsatisfactionofthenetwork.Furthermore,weestablishatestbedusingUSRPtoshowcaseourdesignsinrealscenarios.Tothebestofourknowledge,wearethefirsttoproposeapproximationalgorithmforrelayplacementproblemtomaximizeusersatisfaction,whichhasboththeoreticalandpracticalsignificanceintherelatedarea.
DispersingSocialContentinMobileCrowdthroughOpportunisticContacts LeiZhang(SimonFraserUniversity),FengWang(TheUniversityofMississippi),JiangchuanLiu(SouthChinaAgriculturalUniversity)
![Page 102: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/102.jpg)
102
Crowdsourcedcontentsharinghasbecomeafast-growingactivityintoday'sonlinesocialnetworks,wherecontentsofinterestarecreatedbydiversesourceusersandconveyedoverthenetworkasfriendsviewandreshare.Therapidandboundlesspropagationinamobilecrowdhoweveroftencreatesseverebottlenecksontheserversideandincurssignificantenergyandmonetarycostsonthemobileside,particularlywiththestillexpensive3G/4Gcellularconnections.ThispaperpresentsSoCrowd,anovelframeworkforlarge-scalecontentsharinginamobilecrowdbyexploitingcontacts,i.e.,usershappentomoveclosewithsuchshortrangelowpowercommunicationsasWiFiandbluetoothbeingenabled.Weformulatetheschedulingproblemforsocialcontentpropagationinamobilecrowdwithcontacts,andpresentoptimaldynamicprogrammingsolution,whichfurthermotivatesaseriesofpracticalheuristics.TheeffectivenessofSoCrowdhasbeendemonstratedbyextensivesimulationsdrivenbyrealworldtracesanddatasets.
ALightweightRecommendationFrameworkforMobileUser'sLinkSelectioninDenseNetwork JiWang(NationalUniversityofDefenseTechnology),XiaominZhu(NationalUniversityofDefenseTechnology),WeidongBao(NationalUniversityofDefenseTechnology),GuanlinWu(NationalUniversityofDefenseTechnology)
Withtheproliferationofmobiledevicesandthedevelopmentofcommunicationtechnology,mobiledeviceshavepermeatedeveryaspectofourdailylives.However,indensenetworkwherelargecrowdofmobiledevicestrytoaccesstothenetworksimultaneously,thesevereinterferencebetweenmobiledevicesmayincuraremarkabledeteriorationofthewirelesscommunicationquality.Howtoimproveindividual'sexperienceinsuchscenarioisacriticalyetopenproblem.Inspiredbythemobiledeviceusers'usagepatternaswellasthecharacteristicofmostwirelesscommunicationsystems,weproposeaframeworkofferinguplink/downlinkselectionrecommendationtodifferentmobiledeviceuserstoenhancetheirutilityinthispaper.Thedesignoftheframeworkstartswithformulatingtheproblemasalinkselectiongame.AnalysisshowsthatthegamecanbecategorizedasageneralizedordinalpotentialgamewhoseNashEquilibriumisguaranteed.WethendeviseadistributedlinkselectionalgorithmtogenerateaNashEquilibriumofthegame.Toaccommodatetothecharacteristicofdensenetworkandthecapacitylimitationofmobiledevice,thedesignofthealgorithmshowsalight-weightpropertyanddoesnotrequireeachmobiledeviceusertoknowothers'currentselection.Theprobabilityofincompleteinformationgatheringisalsoconsidered.Extensiveexperimentsareconductedtodemonstratetheeffectivenessandsuperiorityoftheproposedframework.Experimentalresultsshowthattheglobalaverageutilityincreaseratereachesabove20%,andabout70%mobiledeviceuserscanbenefitfromusingourframework
MakingSmartphoneSmartonDemandforLongerBatteryLife MarcoBrocanelli(TheOhioStateUniversity),XiaoruiWang(TheOhioStateUniversity)
Amajorconcernfortoday’ssmartphonesistheirmuchfasterbatterydrainthantraditionalfeaturephones,despitetheirgreaterbatterycapacities.Thedifferenceismainlycontributedbythosemorepowerfulbutalsomuchmorepowerconsumingsmartphonecomponents,suchasthemulti-coreapplicationprocessor.Whiletheapplicationprocessormustbeactivewhenanysmartappsarebeingused,itisalsounnecessarilywakenup,evenduringidleperiods,toperformoperationsrelatedtobasicphonefunctions(i.e.,incomingcallsandtextmessages).Inthispaper,weinvestigatehowtoincreasethebatterylifeofsmartphonesbyminimizingtheuseoftheapplicationprocessorduringidleperiods.Wefindthattheapplicationprocessorisoftenwakenupbyaprocessrunningonit,calledtheRadioInterfaceLayerDaemon(RILD),whichinterfacestheuserandappstotheGSM/LTEcellularnetwork.Inparticular,wedemonstratethatagreatamountofenergycouldbesavedifRILDisstopped,suchthattheapplicationprocessorcansleepmoreoften.Basedonthiskeyfinding,wedesignaSmartOnDemand(SOD)configurationthatreducessmartphoneidleenergyconsumptionbyrunningRILDoperationsonasecondarylow-powermicrocontroller.Asaresult,RILDoperationscanbehandledatmuchlowerenergycostsandtheapplicationprocessoriswakenuponlywhenoneneedstouseanysmartapps,inanon-demandmanner.WehavebuiltahardwareprototypeofSODandevaluateditwithrealusertraces.OurresultsshowthatSODcanincreaseitsbatterylifebyupto2.5moredays.
FADEWICH:FastDeauthenticationovertheWirelessChannel GiulioLovisotto(UniversityofOxford),MauroConti(UniversityofPadua),IvanMartinovic(UniversityofOxford),GeneTsudik(UniversityofCalifornia,Irvine)
Bothauthenticationanddeauthenticationareinstrumentalforpreventingunauthorizedaccesstocomputersandotherresources.Whilethereareobviousmotivatingfactorsforusingstrongauthenticationmechanisms,convincinguserstodeauthenticateisnotstraight-forward,sincedeauthenticationisnotconsideredmandatory.Auserwholeavesalogged-inworkstationunattended(especiallyforashorttime)istypicallynotinconveniencedinanyway;infact,theotherwayaround–noannoyingreauthenticationisneededuponreturn.However,anunattendedworkstationistriviallysusceptibletothewell-known“lunchtimeattack”byanynearbyadversarywhosimplytakesoverthedeparteduser’slog-insession.Atthesametime,sincedeauthenticationdoesnotintrinsicallyrequireusersecrets,itcan,inprinciple,bemadeunobtrusive.Tothisend,thispaperdesignsthefirstautomaticuserdeauthenticationsystem–FADEWICH–thatdoesnotrelyonbiometric-orbehavior-basedtechniques(e.g.,keystrokedynamics)anddoesnotrequireuserstocarryanydevices.Itusesphysicalpropertiesofwirelesssignalsandtheeffectofhumanbodiesontheirpropagation.ToassessFADEWICH’sfeasibilityandperformance,extensiveexperimentswereconductedwithitsprototype.Resultsshowthatitsufficestohavenineinexpensivewirelesssensorsdeployedinasharedofficesettingtocorrectlydeauthenticatealluserswithinsixseconds(90%withinfourseconds)aftertheyleavetheirworkstation’svicinity.WeconsideredtworealisticscenarioswheretheadversaryattemptstosubvertFADEWICHandshowedthatlunchtimeattacksfail.
CognitiveWirelessCharger:Sensing-BasedReal-TimeFrequencyControlForNear-FieldWirelessCharging Sang-YoonChang(UniversityofColoradoColoradoSprings),SristiLakshmiSravanaKumar(AdvancedDigitalSciencesCenter),Yih-ChunHu(UniversityofIllinoisatUrbana-Champaign)
ArecentincreaseinmobileandIoTdeviceshasledtotheadvancementofwirelesscharging.Thestate-of-the-artwirelesschargingsystemsoperateataparticularfrequency,controlledbytheexplicitnetworkingfromthepower-receivingdevice(whichrelaysthebatterystatusinformation,usefulforthefrequencyselection),butsuchcontrolisnotdesignedtocopewiththevariationsinthepowerreceivingdevice’splacementsandalignments(whicharemoresignificantinnear-fieldandpseudo-tightlycoupledchargingapplications,asmorechargingpadsarebeingdeployedinthepublicdomainsandservingheterogeneousclients).Inthiswork,weanalyzetheimpactofthepowertransferperformancecausedbythepowerreceiver’sload,distance,andcoilalignment/overlapandintroducecognitivewirelesscharger(CWC),whichadaptivelycontrolstheoperatingfrequencyinreal-timeusingimplicitfeedbackfromsensingforoptimaloperations.InadditiontothetheoreticalandLTSpice-basedsimulationanalysis,webuildaprototypecompatibletotheQistandardandanalyzetheperformanceofCWCwithit.Throughouranalyses,weestablishthatfrequencycontrolachievesperformancegainsininductive-couplingchargingapplicationsandissensitivetothevariationsintheplacementandalignmentbetweenthepower-transmittingandthepower-receivingcoils.Ourprototype,whenCWCisturnedoff,hascomparableperformancetothecommercial-gradeQiwirelesschargersand,withCWCenabled,demonstratessignificantimprovementovermodernwirelesschargers.
![Page 103: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/103.jpg)
103
DensityandMobility-drivenEvaluationofBroadcastAlgorithmsforMANETs RazielCarvajalGómez(UniversityofNeuchatel),IntiGonzalez-Herrera(LaBRI/UniversityofBordeaux),Yérom-DavidBromberg(UniversityofRennes1),LaurentRéveillère(UniversityofBordeaux),EtienneRivière(UniversityofNeuchatel)
BroadcastisafundamentaloperationinMobileAd-HocNetworks(MANETs).Alargevarietyofbroadcastalgorithmshavebeenproposed.Theydifferinthewaymessageforwardingbetweennodesiscontrolled,andinthelevelofinformationaboutthetopologythatthiscontrolrequires.DeploymentscenariosforMANETsvarywidely,inparticularintermsofnodesdensityandmobility.Thechoiceofanalgorithmdependsonitsexpectedcoverageandenergycost,whicharebothimpactedbythedeploymentcontext.Inthiswork,weareinterestedinthecomprehensivecomparisonofthecostsandeffectivenessofbroadcastalgorithmsforMANETsdependingontargetenvironmentalconditions.Wedescribetheresultsofanexperimentalstudyoffivealgorithms,representativeofthemaindesignalternatives.Ourstudyrevealsthatthebestalgorithmforagivensituation,suchasahighdensityandastablenetwork,isnotnecessarilythemostappropriateforadifferentsituationsuchasasparseandmobilenetwork.Weidentifythealgorithmscharacteristicsthatarecorrelatedwiththesedifferencesanddiscusstheprosandconsofeachdesign.
Energy-AwareCPUFrequencyScalingforMobileVideoStreaming WenjieHu(ThePennsylvaniaStateUniversity),GuohongCao(ThePennsylvaniaStateUniversity)
TheenergyconsumedbyvideostreamingincludestheenergyconsumedfordatatransmissionandCPUprocessing,whicharebothaffectedbytheCPUfrequency.HighCPUfrequencycanreducethedatatransmissiontimebutitconsumesmoreCPUenergy.LowCPUfrequencyreducestheCPUenergybutincreasesthedatatransmissiontimeandthenincreasestheenergyconsumption.Inthispaper,weaimtoreducethetotalenergyofmobilevideostreamingbyadaptivelyadjustingtheCPUfrequency.Basedonrealmeasurementresults,wemodeltheeffectsofCPUfrequencyonTCPthroughputandsystempower.Basedonthesemodels,weproposeanEnergy-awareCPUFrequencyScaling(EFS)algorithmwhichselectstheCPUfrequencythatcanachieveabalancebetweensavingthedatatransmissionenergyandCPUenergy.Sincethedownloadingscheduleofexistingvideostreamingappsisnotoptimizedintermsofenergy,wealsoproposeamethodtodeterminewhenandhowmuchdatatodownload.Throughtrace-drivensimulationsandrealmeasurement,wedemonstratethattheEFSalgorithmcanreduce30%ofenergyfortheYoutubeapp,andthecombinationofourdownloadmethodandEFSalgorithmcansave50%ofenergythanthedefaultYoutubeapp.
CrazyCrowdSourcingtoMitigateResourceScarcity NovaAhmed(NorthSouthUniversity),MdMahfuzurRahmanSiddiquee(IndependentResearcher),RefayaKarim(NorthSouthUniversity),MohsinaZaman(IndependentResearcher),SayedMahmudulAlam(NorthSouthUniversity),SyedFahimAsraf(NorthSouthUniversity)
Resourcescarcityprohibitsdevelopingcountrypopulationinmanywaysfromubiquitousservices.OnesuchserviceisprovidinginformationaboutthebestrouteforanambulanceinacrisissituationduetolackofproperroadnetworkinformationandGPSdata.WehaveworkedonaroutingmethodinDhaka,Bangladeshandutilizedthepowerofcrowdsourcingintimesofresourcescarcity.Weshareourchallengesandopportunitiesthatopenedupfollowedbythechallengesinthispaper.
DetectingRogueAPwiththeCrowdWisdom TongqingZhou(NationalUniversityofDefenseTechnology),ZhipingCai(NationalUniversityofDefenseTechnology),BinXiao(TheHongKongPolytechnicUniversity),YueyueChen(NationalUniversityofDefenseTechnology),MingXu(NationalUniversityofDefenseTechnology)
WiFinetworksarevulnerabletorogueAPattacksinwhichanattackersetsupanimposterAPtoluremobileuserstoconnect.Theattackercaneavesdroponthecommunication,severelythreateningusers'privacy.ExistingrogueAPdetectionsolutionsareconfinedtosomespecificattackscenarios(e.g.,byrelayingthetraffictoatargetAP)orrequireadditionalhardware.Inthispaper,weproposeacrowdsensingbasedapproach,namedCRAD,todetectrogueAPsincamouflagewithoutspecializedhardwarerequirement.CRADexploitsthespatialcorrelationofRSStoidentifyapotentialimposter,whichshouldbeatadifferentlocationfromthelegitimateone.TheRSSmeasurementscollectedfromthecrowdfacilitatearobustprofileandminimizetheinaccuracyeffectofasingleRSSvalue.Asaresult,CRADcanfilteroutabnormalsamplessensedintherealtimebydynamicallymatchingtheprofile.Weevaluateourapproachwithbothapublicdatasetandarealprototype.TheresultsshowthatCRADcanyield90%detectionaccuracyandprecisionwithpropercrowdpresence,evenwhentherogueAPislaunchedclosetothelegitimateone(e.g.,within1m).
ShortPaper5:DistributedBigDataSystemsandAnalyticsTowardsMultilingualAutomatedClassificationSystems AibekMusaev(UniversityofAlabama),CaltonPu(GeorgiaInstituteofTechnology)
Inthispaperweproposeandevaluatethreeapproachesforautomatedclassificationoftextsinover60languageswithouttheneedforamanuallyannotateddatasetinthoselanguages.AllapproachesarebasedontherandomizedExplicitSemanticAnalysismethodusingmultilingualWikipediaarticlesastheirknowledgerepository.WeevaluatetheproposedapproachesbyclassifyingaTwitterdatasetinEnglishandPortugueseintorelevantandirrelevantitemswithrespecttolandslideasanaturaldisaster,wherethehighestachievedF1-scoreis0.93.Theseapproachescanbeusedinvariousapplicationswheremultilingualclassificationisneeded,includingmultilingualdisasterreportingusingSocialMediatoimprovecoverageandincreaseconfidence.Asillustration,wepresentademonstrationthatcombinesdatafromphysicalsensorsandsocialnetworkstodetectlandslideeventsreportedinEnglishandPortuguese.
TheJointEffectsofTweetContentSimilarityandTweetInteractionsforTopicDerivation RobertusNugroho(MacquarieUniversity),WeiliangZhao(MacquarieUniversity),JianYang(MacquarieUniversity),CecileParis(CSIRO–ICTCentre),SuryaNepal(CSIRO)
Interactionsamongtweets,i.e.,mentions,retweets,replies,areimportantfactorscontributingtothequalityoftopicderivationonTwitter.Ifappliedcorrectly,theincorporationoftweetinteractionscansignificantlyimprovethequalityoftopicderivationincomparisonwithapproachesthataremainlybasedonthecontentsimilarityanalysis.However,howinteractionscanbemeasuredandintegratedwithcontentsimilarityfortopicderivationremainsachallenge.Inpreviouswork,thestrengthoftweet-to-tweetrelationshiphasbeencomputedbysimplyaddingmeasuresforcontentsimilarity,mentions,andreply-retweets.Thissimplelinearadditiondoesnotaccuratelyreflectthevariousimpactsthesefactorshaveontweetrelationships.Inordertoaddressthisissue,weproposeajointprobabilitymodelthatcaneffectivelyintegratetheeffectsofthecontentsimilarity,mentions,andreply-retweetstomeasurethetweetrelationshipforthepurposeoftopic
![Page 104: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/104.jpg)
104
derivation.Theproposedmethodisbasedonmatrixfactorizationtechniques,whichenablesaflexibleimplementationonadistributedsysteminanincrementalmanner.Experimentalresultsshowthattheproposedmodelresultsinasignificantimprovementinthequalityoftopicderivationoverexistingmethods.
Timed-releaseofSelf-emergingDatausingDistributedHashTables ChaoLi(UniversityofPittsburgh),BalajiPalanisamy(UniversityofPittsburgh)
Releasingprivatedatatothefutureisachallengingproblem.Makingprivatedataaccessibleatafuturepointintimerequiresmechanismstokeepdatasecureandundiscoveredsothatprotecteddataisnotavailablepriortothelegitimatereleasetimeandthedataappearsautomaticallyattheexpectedreleasetime.Inthispaper,wedevelopnewmechanismstosupportself-emergingdatastoragethatsecurelyhidekeysofencrypteddatainaDistributedHashTable(DHT)networkthatmakestheencryptionkeysautomaticallyappearatthepredeterminedreleasetimesothattheprotectedencryptedprivatedatacanbedecryptedatthereleasetime.Weshowthatastraight-forwardapproachofprivatelystoringkeysinaDHTispronetoanumberofattacksthatcouldeithermakethehiddendataappearbeforetheprescribedreleasetime(release-aheadattack)ordestroythehiddendataaltogether(dropattack).Wedevelopasuiteofself-emergingkeyroutingmechanismsforsecurelystoringandroutingencryptionkeysintheDHT.Weshowthattheproposedschemeisresilienttobothrelease-aheadattackanddropattackaswellastoattacksthatariseduetotraditionalchurnissuesinDHTnetworks.Ourexperimentalevaluationdemonstratestheperformanceoftheproposedschemesintermsofattackresilienceandchurnresilience.
CachingforPatternMatchingQueriesinTimeEvolvingGraphs:ChallengesandApproaches MuhammadNisar(UniversityofGeorgia),SaharVoghoei(UniversityofGeorgia),LakshmishRamaswamy(UniversityofGeorgia)
Patternmatchingisanimportantclassofproblemsrelatedtographs.Itisafundamentalproblemformanyapplicationsandhasbeenextensivelystudiedinliterature.Withtheadventofhugegraphs,thechallengesinthisdomainhaveincreasedmanifold.Consequentlyalotofrecentresearchhasledtonewarchitecturesandapproachesforoptimizedsolutionstothepatternmatchingproblem.Avastmajorityofthesegraphshardlyremainstaticandconstantlyevolveovertime(likesocialnetworks,webgraphs,etc).Recently,cachinghasbeenstudiedinthecontextofstaticgraphstooptimizethethroughputofqueryprocessingsystems.Inthispaper,welistthechallengesincachinginthecontextofTimeEvolvingGraphs(TEGs).Amongstothers,onemajorchallengeisconsistencywhichentailstomakingsurethecacheisconsistentwiththestreamingchanges.Weproposeanapproachtosuccessfullyimplementcachingthataddressesthoseissuesandbasedontheinitialresults,weseesignificantgainsintheoverallperformanceofsystem.
GraphA:AdaptivePartitioningforNaturalGraphs DongshengLi(NationalUniversityofDefenseTechnology),ChengfeiZhang(NationalUniversityofDefenseTechnology),JinyanWang(NationalUniversityofDefenseTechnology),ZhaoningZhang(NationalUniversityofDefenseTechnology),YimingZhang(NationalUniversityofDefenseTechnology)
Large-scalegraphcomputationiscentraltoapplicationsrangingfromlanguageprocessingtosocialnetworks.However,naturalgraphstendtohaveskewedpower-lawdistributionswhereasmallsubsetoftheverticeshavealargenumberofneighbors.Existinggraph-parallelsystemssufferfromloadimbalance,highcommunicationcost,orsuboptimalandcomplexprocessing.InthispaperwepresentGraphA,anAdaptiveapproachtoefficientpartitioningandcomputationoflarge-scalenaturalgraphs.GraphAprovidesanadaptiveanduniformgraphpartitioningalgorithm,whichpartitionsthedatasetsinaload-balancedmannerbyusinganincrementalnumberofhashfunctions.WehaveimplementedGraphAbothonSparkandonGraphLab.ExtensiveevaluationshowsthatGraphAremarkablyoutperformsstate-of-the-artgraph-parallelsystems(GraphXandPowerLyra)iningresstime,executiontimeandstorageoverhead,forbothreal-worldandsyntheticgraphs.
ParallelAlgorithmforCoreMaintenanceinDynamicGraphs NaWang(HuazhongUniversityofScienceandTechnology),DongxiaoYu(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),ChenQian(HuazhongUniversityofScienceandTechnology),XiaXie(HuazhongUniversityofScienceandTechnology),Qiang-ShengHua(HuazhongUniversityofScienceandTechnology)
Thispaperinitiatesthestudiesofparallelalgorithmforcoremaintenanceindynamicgraphs.Thecorenumberisafundamentalindexreflectingthecohesivenessofagraph,whichiswidelyusedinlarge-scalegraphanalytics.Weinvestigatetheparallelisminthecoreupdateprocesswhenmultipleedgesandverticesareinserted.Specifically,wediscoverastructurecalledsuperioredgeset,theinsertionofedgesinwhichcanbeprocessedinparallel.Basedonthestructureofsuperioredgeset,anefficientparallelalgorithmisthendevised.Tothebestofourknowledge,theproposedalgorithmisthefirstparalleloneforthefundamentalcoremaintenanceproblem.Finally,extensiveexperimentsareconductedondifferenttypesofreal-worldandsyntheticdatasets,andtheresultsillustratetheefficiency,stabilityandscalabilityoftheproposedalgorithm.Thealgorithmshowsasignificantspeedupintheprocessingtimecomparedwithpreviousresultsthatsequentiallyhandleedgeandvertexinsertions.
DHCRF:ADistributedConditionalRandomFieldAlgorithmonHeterogeneousCPU-GPUClusterforBigData AiWei(HunanUniversity),LiKenli(HunanUniversity),ChenCen(HunanUniversity),PengJiwu(HunanUniversity),LiKeqin(HunanUniversity)
Asoneofthemostrecognizedmodelsinmachinelearning,theconditionalrandomfields(CRF)hasbeenwidelyusedinmanyapplications.AstheparameterestimationofCRFishighlytime-consuming,howtoimprovetheperformanceofCRFhasreceivedsignificantattention,inparticularinthebigdataenvironment.Todealwithlarge-scaledata,CPU-basedorGPU-basedparallelizationsolutionshavebeenproposedtoimproveperformance.However,theproblemisanongoingone.Inthispaper,wefocusonthebigdataenvironmentandproposeadistributedCRFonaheterogeneousCPU-GPUclustercalledDHCRF.Ourapproachdiffersfrompreviouswork.Specifically,itleveragesathree-stageheterogeneousMapandReduceoperationtoimprovetheperformance,makingfulluseofCPU-GPUcollaborativecomputingcapabilitiesinabigdataenvironment.Furthermore,bycombiningelasticdatapartitionandintermediateresultsmultiplexingmethod,thedistributedCRFisoptimized.Elasticdatapartitionisperformedtokeeptheloadbalanced,andtheintermediateresultsmultiplexingmethodisadoptedtoreducedatacommunication.ExperimentalresultsshowthattheDHCRFoutperformsthebaselineCRFalgorithmandtheCPU-basedparallelCRFalgorithmwithnotableperformanceimprovementwhilemaintainingcompetitivecorrectnessatthesametime.
TowardsNewAbstractionsforImplementingQuorum-basedSystems TormodErevikLea(UniversityofStavanger),LeanderJehl(UniversityofStavanger),HeinMeling(UniversityofStavanger)
![Page 105: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/105.jpg)
105
ThispaperintroducesGorums,anovelRPCframeworkforbuildingfaulttolerantdistributedsystems.Gorumsoffersaflexibleandsimplequorumcallabstraction,usedtocommunicatewithasetofprocesses,andtocollectandprocesstheirresponses.Gorumsprovidesseparateabstractionsfor(a)selectingprocessesforaquorumcalland(b)processingreplies.Theseabstractionssimplifythemaincontrolflowofprotocolimplementations,especiallyforquorum-basedsystems,whereonlyasubsetoftherepliestoaquorumcallneedtobeprocessed.ToshowthatGorumscanbeusedinpracticalsystems,weimplementedEPaxos’latency-efficientquorumsystem,andranexperimentsusingakey-valuestorage.OurresultsshowthatGorums’abstractionscanprovideadditionalperformancebenefitstoEPaxos.
SelectiveTrafficOffloadingOntheFly:aMachineLearningApproach ZaiyangTang(HuazhongUniversityofScienceandTechnology),PengLi(TheUniversityofAizu),SongGuo(TheHongKongPolytechnicUniversity),XiaofeiLiao(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology),DaqingZhang(InstitutMines-Telecom,TelecomSudParis)
Ithasbeenwellrecognizedthatnetworktransmissionconstitutesalargeportionofsmartphoneenergyconsumption,mainlybecauseofthetailenergycausedbycellularnetworkinterface.Trafficoffloadinghasbeenproposedtoreduceenergybylettingasmartphoneoffloadnetworktraffictoitsneighborsinvicinityvialow-powerdirectconnections(e.g.,WiFiDirectorBluetooth).Ourexperimentsconductedinarealisticenvironmentrevealthatenergyefficiencycannotbeimprovedorevendeteriorateswithoutacarefullydesignedoffloadingstrategy.Inthispaper,weproposeaselectivetrafficoffloadingschemeimplementedasasmartphonemiddlewareinasoftware-definedfashion,whichconsistsofapacketclassifierandatrafficscheduler.Usingalight-weightmachinelearningapproachexploitinguniquesmartphonecontextinformation,thepacketclassifieridentifiespacketsgeneratedontheflyasoffloadableornotwithsubstantiallyimprovedefficiencyandfeasibilityonresourcelimitedsmartphonescomparedtotraditionalapproaches.Bothtestbedandsimulationbasedexperimentsareconductedandtheresultsshowthatourproposalalwaysattainsthesuperiorperformanceonanumberofcomparisonmetrics.
AFastHeuristicAttributeReductionAlgorithmusingSpark MinchengChen(WuhanUniversityofTechnology),JinglingYuan(WuhanUniversityofTechnology),LinLi(WuhanUniversityofTechnology),DonglingLiu(WuhanUniversityofTechnology),TaoLi(UniversityofFlorida)
Energydata,whichconsistsofenergyconsumptionstatisticsandotherrelateddataingreendatacenters,growsdramatically.Theenergydatahasgreatvalue,butmanyattributeswithinitareredundantandunnecessary.Thusattributereductionfortheenergydatahasbeenconceivedasacriticalstep.However,manyexistingattributereductionalgorithmsareoftencomputationallytime-consuming.Toaddresstheseissues,weextendthemethodologyofroughsetstoconstructdatacenterenergyconsumptionknowledgerepresentationsystem.Bytakinggoodadvantageofin-memorycomputing,anattributereductionalgorithmforenergydatausingSparkisproposed.Inthisalgorithm,weuseaheuristicformulaformeasuringthesignificanceofattributetoreducesearchspace,andanefficientalgorithmforsimplifyingenergyconsumptiondecisiontable,whichfurtherimprovethecomputationefficiency.Theexperimentalresultsshowthespeedofouralgorithmgainsupto0.28XperformanceimprovementoverthetraditionalattributereductionalgorithmusingSpark.
ProfilingUsersbyModelingWebTransactions RadekTomsu(AaltoUniversity),SamuelMarchal(AaltoUniversity),N.Asokan(AaltoUniversity)
Usersofelectronicdevices,e.g.,laptop,smartphone,etc.havecharacteristicbehaviorswhilesurfingtheWeb.Profilingthisbehaviorcanhelpidentifythepersonusingagivendevice.Inthispaper,weintroduceatechniquetoprofileusersbasedontheirwebtransactions.Wecomputeseveralfeaturesextractedfromasequenceofwebtransactionsandusethemwithone-classclassificationtechniquestoprofileauser.Weassesstheefficacyandspeedofourmethodatdifferentiating25syntheticusersonabenchmarkdataset(fromamajorsecurityvendor)representing6monthsofwebtrafficmonitoringfromasmallenterprisenetwork.
JeCache:Just-EnoughDataCachingwithJust-in-TimePrefetchingforBigDataApplications YifengLuo(FudanUniversity),JiaShi(FudanUniversity),ShuigengZhou(FudanUniversity)
Bigdataclustersintroduceanintermediatecachelayerbetweenthecomputingframeworksandtheunderlyingdistributedfilesystems,toenableupper-levelapplicationsorenduserstoefficientlyaccessbigdatasetsincacheandeffectivelysharethemamongdifferentcomputingframeworks.Ascachesaresharedbymultipleapplicationsorendusers,directlyapplyingexistingon-demandcachingstrategieswillresultinintenseconflicts,whenbigdatasetsarecachedasawhole.Meanwhile,bigdataapplicationsusuallyinvolvemassivenumbersoffilescans,cached-indatablocksmayhavelittlechanceofbeingaccessedbeforetheyarecachedouttomakewayforotheron-demanddatablocks.Thus,itisunwisetocachedatablockslongbeforetheyareactuallyaccessed.Inthispaper,weproposeanoveljust-enoughbigdatacachingschemeforjust-in-timeblockprefetchingtoimprovethecacheeffectivenessofbigdataclusters.Withjust-in-timeblockprefetching,ablockiscachedinjustbeforethetaskbeginstoprocesstheblock,ratherthanbeingcachedinalongwithotherblocksofthesamedatasetbeingprocessed.Wemonitorblockaccessestomeasuretheaverageprocessingtimeofdatablocks,andthenestimatetheminimalnumberofblocksthatshouldbekeptincacheforabigdataset,sothatthespeedofdataprocessingmatcheswiththatofdataprefetching,andeachupper-leveltaskcanobtainitsinputblocksfromcachejustintime.Ourexperimentalresultsshowthattheproposedcachemethodcanrestrainover-requirementofcacheresourcesinbigdataapplications,andprovidesthesameperformanceimprovementaswhenalldatablocksarecached.
ShortPaper6:Security,Privacy,Trust,andFaultToleranceinDistributedSystemsProximityAwarenessApproachtoEnhancePropagationDelayontheBitcoinPeer-to-PeerNetwork MuntadherFadhilSallal(UniversityofPortsmouth),GarethOwenson(UniversityofPortsmouth),MoAdda(UniversityofPortsmouth)
IntheBitcoinsystem,apeer-to-peerelectroniccurrencysystem,thedelayoverheadintransactionverificationpreventstheBitcoinfromgainingincreasingpopularitynowadaysasitmakesthesystemvulnerabletodoublespendattacks.Thispaperintroducesaproximity-awareextensiontothecurrentBitcoinprotocol,namedBitcoinClusteringBasedPingTimeprotocol(BCBPT).Theultimatepurposeoftheproposedprotocol,thatisbasedonhowtheclustersareformulatedandthenodesdefinetheirmembership,istoimprovethetransactionpropagationdelayintheBitcoinnetwork.InBCBPT,theproximityofconnectivityintheBitcoinnetworkisincreasedbygroupingBitcoinnodesbasedonpinglatenciesbetweennodes.Weshow,throughsimulations,thattheproximitybasepinglatencydefinesbetterclusteringstructuresthatoptimizetheperformanceofthetransactionpropagationdelay.Thereductionofthecommunicationlinkcostmeasuredbytheinformationpropagationtimebetweennodesismainlyconsideredasakeyreasonforthisimprovement.BitcoinClusteringBasedPingTimeprotocolismoreeffectiveatreducingthetransactionpropagationdelaycomparedtotheexistingclusteringprotocol(LBC)thatweproposedinourpreviouswork.
![Page 106: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/106.jpg)
106
CatchMeIfYouCan:DetectingCompromisedUsersThroughPartialObservationonNetworksDerekWang(DeakinUniversity),ShengWen(DeakinUniversity),JunZhang(DeakinUniversity),SuryaNepal(Data61),YangXiang(DeakinUniversity),WanleiZhou(DeakinUniversity)
Peoplearesufferingfromarangeofrisksintheubiquitousnetworksofcurrentworld,suchasrumoursspreadinginsocialnetworks,computervirusespropagatingthroughouttheInternetandunexpectedfailureshappenedinSmartgrids.Weusuallymonitoronlyafewusersofdetectingvariousrisksduetotheresourceconstraintsandprivacyprotection.Thisleadstoacriticalproblemtodetectcompromiseduserswhoareoutofsurveillance.Inthispaper,weproposeariskassessmentmethodtoaddressthisproblem.Theaimistoassessthesecuritystatusofunmonitoredusersaccordingtothelimitedinformationcollectedfrommonitoredusersinnetworks.Therearetwoinnovativetechniquesdeveloped:First,weidentifythesourceofriskpropagationbyinverselydisseminatingrisksfromtheinfluenced(byrumours)orinfected(byviruses)monitoredusers.Weshowanewfindingthattheoneswhosynchronouslyreceivetheriskcopiesfromallmonitoredusersaremostlikelytobethesources.Second,weproposeamicroscopicmathematicalmodeltopresenttheriskpropagationfromtheexposedsources.Thismodelformsadiscriminanttoclassifythecompromisedusersfromothers.Forevaluations,wecollectthreerealnetworksonwhichwelaunchsimulatedriskpropagationandthensamplethestatusofmonitoredusers.Theexperimentresultsshowthatourmethodiseffectiveandtheresultofriskassessmentwellmatchestherealstatusoftheunmonitoredusers.
LocationPrivacyBreach:AppsAreWatchingYouinBackground DachuanLiu(CollegeofWilliam&Mary),XingGao(CollegeofWilliam&Mary),HainingWang(UniversityofDelaware)
SmartphoneuserscanconvenientlyinstallasetofappsthatprovideLocationBasedService(LBS)frommarkets.TheseLBS-basedappsfacilitateusersinmanyapplicationscenarios,buttheyraiseconcernsonthebreachofprivacyrelatedtolocationaccess.Smartphoneuserscanhardlyperceivelocationaccess,especiallywhenithappensinbackground.Incomparisontolocationaccessinforeground,locationaccessinbackgroundcouldresultinmoreseriousprivacybreachbecauseitcancontinuouslyknowauser’slocations.Inthispaper,westudytheproblemoflocationaccessinbackground,andespeciallyperformthefirstmeasurementofthisbackgroundactionontheGoogleappmarket.Ourinvestigationdemonstratesthatmanypopularappsconductlocationaccessinbackgroundwithinshortintervals.Thisenablestheseappstocollectauser’slocationtrace,fromwhichtheimportantpersonalinformation,PointsofInterest(PoIs),canberecognized.Wefurtherextractauser’smovementpatternfromthePoIs,andutilizeittomeasuretheextentofprivacybreach.ThemeasurementresultsalsoshowthatusingthecombinationofmovementpatternrelatedmetricsandtheotherPoIrelatedmetricscanhelpdetecttheprivacybreachinanearliermannerthanusingeitheroneofthemalone.
AndroidMalwareDetectionusingComplex-Flows FengShen(SUNYBuffalo),JustinDelVecchio(SUNYBuffalo),AzizMohaisen(SUNYBuffalo),StevenY.Ko(SUNYBuffalo),LukaszZiarek(SUNYBuffalo)
Thispaperproposesanewtechniquetodetectmobilemalwarebasedoninformationflowanalysis.Ourapproachexaminesthestructureofinformationflowstoidentifypatternsofbehaviorpresentinthemandwhichflowsarerelated,thosethatsharepartialcomputationpaths.WecallsuchflowsComplex-Flows,astheirstructure,patterns,andrelationsaccuratelycapturethecomplexbehaviorexhibitedbybothrecentmalwareandbenignapplications.N-gramanalysisisusedtoidentifyuniqueandcommonbehavioralpatternspresentinComplex-Flows.TheN-gramanalysisisperformedonsequencesofAPIcallsthatoccuralongComplex-Flows'controlflowpaths.Weshowtheprecisionofourtechniquebyapplyingittodifferentdatasetstotaling7,798apps.Thesedatasetsconsistofbothrecentandoldergenerationbenignandmaliciousappstodemonstratetheeffectivenessofourapproachacrossdifferentgenerationsofapps.
PrivacyImplicationsofDNSSECLook-asideValidation AzizMohaisen(SUNYBuffalo),ZhongshuGu(IBMResearch),KuiRen(SUNYBuffalo)
TocomplementDNSSECoperations,DNSSECLook-asideValidation(DLV)isdesignedforalternativeoff-pathvalidation.WhileDNSprivacyattractsalotofattention,theprivacyimplicationsofDLVarenotfullyinvestigatedandunderstood.Inthispaper,wetakeafirstin-depthlookintoDLV,highlightingitslaxspecificationsandprivacyimplications.Byperformingextensiveexperimentsoverdatasetsofdomainnamesundercomprehensiveexperimentalsettings,ourfindingsfirmlyconfirmtheprivacyleakagescausedbyDLV.WediscoverthatalargenumberofdomainsthatshouldnotbesenttoDLVserversarebeingleaked.Weexploretherootcauses,includingthelaxspecificationsofDLV.Wealsoproposetwoapproachestofixtheprivacyleakages.OurapproachesrequiretrivialmodificationstotheexistingDNSstandardsandwedemonstratetheircostintermsoflatencyandcommunication.
FlipNet:ModelingCovertandPersistentAttacksonNetworkedResources SudipSaha(VirginiaPolytechnicInstituteandStateUniversity),AnilVullikanti(VirginiaPolytechnicInstituteandStateUniversity),MahanteshHalappanavar(PacificNorthwestNationalLab)
Persistentandzero-dayattackshaveincreasedconsiderablyintherecentpastintermsofscaleandimpact.Securityexpertscannolongerrelyonlyonknowndefensesandtherebyprotecttheirresourcespermanently.Itisincreasinglycommonnowtoobserveattackersbeingabletorepeatedlybreaksystemsexploitingnewvulnerabilitiesanddefendershardeningsystemswithnewmeasures.Tomodelthisphenomenonoftherepeatedtakeoverofthecomputingresourcesbysystemadministratorsandmaliciousattackers,anovelgameframework,FlipIt,hasbeenproposedby(VanDijketal.2013)forasystemconsistingofasingleresource.Inthispaper,weextendthisanddevelopFlipNet,whichisarepeatedgameframeworkforanetworkedsystemofmultipleresources.Thisgameinvolvestwoplayers---adefenderandanattacker.Eachplayer'sobjectiveistomaximizeitsgain(i.e.,itscontroloverthenodesinthenetworkwithstealthymoves),whileminimizingthecostformakingthosemoves.Thisleadstoanovelandnaturalgameformulation,withaverycomplexstrategyspace,thatdependsonthenetworkstructure.WeshowthatfindingthebestresponsestrategyforboththedefenderandattackerisNP-hard.Inakeyresultinthisstudy,weshowthattheattacker'sgainforaninstanceofthegamehasatypeofdiminishingmarginalreturnproperty,whichleadstoanear-optimalalgorithmformaximizingtheattacker'sgain.Weexaminetheimpactofnetworkstructureonthestrategyspaceusingsimulations.
UnderstandingtheMarket-levelandNetwork-levelBehaviorsoftheAndroidMalwareEcosystem ChaoYang(Niara,Inc.),JialongZhang(IBMResearch),GuofeiGu(TexasA&MUniversity)
TheprevalenceofmalwareinAndroidmarketplacesisagrowingandsignificantproblem.MostexistingstudiesfocusondetectingAndroidmalwareordesigningnewsecurityextensionstodefendagainstspecifictypesofattacks.Inthispaper,weperformanempiricalstudyonanalyzingthemarket-levelandnetwork-levelbehaviorsoftheAndroidmalwareecosystem.Wefocusonstudyingwhetherthereareinterestingcharacteristicsofthosemarketaccountsthatdistributemalware
![Page 107: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/107.jpg)
107
andspecificnetworksthataremainlyutilizedbyAndroidmalwareauthors.WefurtherinvestigatecommunitypatternsamongAndroidmalwarefromtheperspectiveoftheirmarketaccountinfrastructureandremoteserverinfrastructure.Spurredbytheseanalysis,wedesignanovelcommunityinferencealgorithmtofindmoremaliciousappsbyexploitingtheircommunityrelationships.Byusingasmallseedset(50)ofknownmaliciousapps,wecaneffectivelyfindanotherextra20timesofmaliciousapps,whilemaintainingconsiderableaccuracyhigherthan94%.
EnGarde:Mutually-TrustedInspectionofSGXEnclaves. HaiNguyen(RutgersUniversity),VinodGanapathy(RutgersUniversity)
Intel’sSGXarchitectureallowscloudclientstocreateenclaves,whosecontentsarecryptographicallyprotectedbythehardwareevenfromthecloudprovider.Whilethisfeatureprotectstheconfidentialityandintegrityoftheclient’senclavecontent,italsomeansthatenclavecontentiscompletelyopaquetothecloudprovider.Thus,thecloudproviderisunabletoenforcepolicycomplianceonenclaves.Inthispaper,weintroduceEnGarde,asystemthatallowscloudproviderstoensureSLAcomplianceonenclavecontent.InEnGarde,cloudprovidersandclientsmutuallyagreeuponasetofpoliciesthattheclient’senclavecontentmustsatisfy.EnGardeexecuteswhentheclientprovisionstheenclave,ensuringthatonlypolicy-compliantcontentisloadedintotheenclave.EnGardeisabletoachieveitsgoalswithoutcompromisingthesecurityguaranteesofferedbytheSGX,andimposesnoruntimeoverheadontheexecutionofenclavecode.WehavedemonstratedtheutilityofEnGardebyusingittoenforceavarietyofsecuritypoliciesonenclavecontent.
TruthfulOnlineAuctionforCloudInstanceSubletting YifeiZhu(SimonFraserUniversity),SilveryFu(SimonFraserUniversity),JiangchuanLiu(SimonFraserUniversity),YongCui(TsinghuaUniversity)
DespitethatIaaSusersarebusyscalingup/outtheircloudinstancestomeettheever-increasingdemands,thedynamicsoftheirdemands,aswellasthecoarse-grainedbillingoptionsofferedbyleadingcloudproviders,haveledtosubstantialinstanceunderutilizationinbothtemporalandspatialdomains.Thispapertheoreticallyexaminesaninstancesublettingservice,whereunderutilizedinstancesareleasedtootherswithinuser-specifiedperiods.ServingasasecondarymarketthatcomplementstheexistinginstancemarketofIaaSproviders,wespecificallyidentifythetheoreticalchallengesininstancesublettingservices,anddesignanonlineauctionmechanismtomakeallocationandpricingdecisionsfortheinstancestobesublet.Ourmechanismguaranteestruthfulnessandindividualrationalitywiththebestpossiblecompetitiveratio.Extensivetrace-drivensimulationsshowthatourproposedmechanismachievessignificantperformancegainsinbothcostandsocialwelfare.
OnthePowerofWeakerPairwiseInteraction:Fault-TolerantSimulationofPopulationProtocols GiuseppeAntonioDiLuna(LaSapienza),PaolaFlocchini(UniversityofOttawa),TaisukeIzumi(NagoyaInstituteofTechnology),TomokoIzumi(CollegeofInformationScienceandEngineering),NicolaSantoro(CarletonUniversity),GiovanniViglietta(UniversityofOttawa)
Inthispaperweinvestigatethecomputationalpowerofpopulationprotocolsundersomeunreliableorweakerinteractionmodels.Moreprecisely,wefocusontwofeaturesrelatedtothepowerofinteractions:omissionfailuresandone-waycommunications.Westartourinvestigationbyprovidingacompleteclassificationofallthepossiblemodelsarisingfromtheaforementionedweaknesses,andestablishingthecomputationalhierarchyofthesemodels.Wethenaddressforeachmodelthefundamentalquestionofwhatadditionalpowerisnecessaryandsufficienttocompletelyovercomethemodel’sweaknessandmakeitabletosimulatefaultlesstwo-wayprotocols.Weanswerthisquestionbypresentingsimulatorsthatworkundercertainassumptionsandbyprovingthatsimulationisimpossiblewithoutsuchassumptions.
DistributedFaultTolerantLinearSystemSolversbasedonErasureCoding XuejiaoKang(PurdueUniversity--WestLafayette),DavidF.Gleich(PurdueUniversity--WestLafayette),AhmedSameh(PurdueUniversity--WestLafayette),AnanthGrama(PurdueUniversity--WestLafayette)
Wepresentefficientcodingschemesanddistributedimplementationsoferasurecodedlinearsystemsolvers.Erasurecodedcomputationsbelongtotheclassofalgorithmicfaulttoleranceschemes.Theyarebasedonaugmentinganinputdataset,executingthealgorithmontheaugmenteddataset,andintheeventofafault,recoveringthesolutionfromthecorrespondingaugmentedsolution.Thisprocesscanbeviewedasthecomputationalanalogoferasurecodedstorageschemes.Theproposedtechniquehasanumberofimportantbenefits:(i)asthehardwareplatformscalesinsizeandnumberoffaults,ourschemeyieldsincreasingimprovementinresourceutilization,comparedtotraditionalschemes;(ii)theproposedschemeiseasytocode–thecorealgorithmsremainthesame;and(iii)thegeneralschemeisflexible–accommodatingarangeofcomputationandcommunicationtradeoffs.Wepresentnewcodingschemesforaugmentingtheinputmatrixthatsatisfytherecoveryequationsoferasurecodingwithhighprobabilityintheeventofrandomfailures.Thesecodingschemesalsominimizefill(non-zeroelementsintroducedbythecodingblock),whilebeingamenabletoefficientpartitioningacrossprocessingnodes.Wedemonstrateexperimentallythatourschemeaddsminimaloverheadforfaulttolerance,yieldsexcellentparallelefficiencyandscalability,andisrobusttodifferentfaultarrivalmodels.
PreservingIncumbentUsers’PrivacyinExclusion-Zone-BasedSpectrumAccessSystems YanzhiDou(VirginiaTech),HeLi(VirginiaTech),KexiongZeng(VirginiaTech),JinshanLiu(VirginiaTech),YalingYang(VirginiaTech),BoGao(ChineseAcademyofSciences),KuiRen(SUNYBuffalo)
Dynamicspectrumaccess(DSA)techniquehasemergedasafundamentalapproachtomitigatethespectrumscarcityproblem.AsakeyformofDSA,thegovernmentisproposingtoreleasemorefederalspectrumforsharingwithcommercialwirelessusers.However,theflourishoffederalcommercialsharinghingesuponhowthefederalprivacyismanaged.IncurrentDSAproposals,thesensitiveexclusionzone(E-Zone)informationoffederalincumbentusers(IUs)needstobesharedwithaspectrumaccesssystem(SAS)torealizespectrumallocation.However,SASisnotnecessarilytrust-worthyforholdingthesensitiveIUE-Zonedata,especiallyconsideringthatFCCallowssomeindustrythirdparties(e.g.,Google)tooperateSASforbetterefficiencyandscalability.Therefore,thecurrentproposalsdissatisfytheIUs’privacyrequirement.Toaddresstheprivacyissue,thispaperpresentsanIU-privacypreservingSAS(IP-SAS)design,whichrealizesthespectrumallocationprocessthroughsecurecomputationoverciphertextbasedonhomomorphicencryptionsothatnoneoftheIUEZoneinformationisexposedtoSAS.ThispaperalsoproposesmechanismstopreventmaliciouspartiesfromcompromisingIP-SAS.Weprovetheprivacy-preservingpropertiesofIP-SASanddemonstratethescalabilityandpracticalityofIP-SASusingexperimentsbasedonreal-worlddata.EvaluationresultsshowthatIP-SAScanrespondanSU’sspectrumrequestin1.25secondswithcommunicationoverheadof17.8KB.
![Page 108: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/108.jpg)
108
Demonstration Track Paper Abstracts
Demo1:DistributedApplicationsCluster
LITMUS:TowardsMultilingualReportingofLandslides AibekMusaev(UniversityofAlabama),QixuanHou(GeorgiaInstituteofTechnology),YangYang(GeorgiaInstituteofTechnology),CaltonPu(GeorgiaInstituteofTechnology)
LITMUSisareal-timeonlineandopenlyaccessibleservicethatcollectshighqualityinformationonlandslideeventsfromsocialmedia.Thisserviceusesdisasterrelatedkeywords,suchas"landslide"and"mudslide",toanalyzemessagespostedbyEnglishspeakingusers.However,comprehensivecoverageofdisastersmustincludemultilingualsupportasthereareeventsthatarereportedinlanguagesotherthanEnglish.Wediscussandevaluatepossibleimplementationsofsuchsupportusing"native"and"translated"approaches."Native"approachinvolvesacompletereimplementationoftheexistinginfrastructureinanotherlanguagewhereasinthe"translated"approachtheexistinginfrastructurecanbeusedwithoutmodification.Asanillustration,wepresentademothatextendsLITMUStoimplementa"native"approachformultilingualreportingoflandslideevents.
Pythia:ASystemforOnlineTopicDiscoveryofSocialMediaPosts IoulianaLitou(AthensUniversityofEconomicsandBusiness),VanaKalogeraki(AthensUniversityofEconomicsandBusiness)
Socialmediaconstitutenowadaysoneofthemostcommoncommunicationmediums.Millionsofusersexploitthemdailytoshareinformationwiththeircommunityinthenetworkviamessages,referredas\emph{posts}.Themassivevolumeofinformationsharedisextremelydiverseandcoversavastspectrumoftopicsandinterests.Automaticallyidentifyingthetopicsofthepostsisofparticularinterestasthiscanassistinavarietyofapplications,suchaseventdetection,trendsdiscovery,expertfindingetc.However,designinganautomatedsystemthatrequiresnohumanagentparticipationtoidentifythetopicscoveredinpostspublishedinOnlineSocialNetworks(OSNs)presentsmanifoldchallenges.First,postsareunstructuredandcommonlyshort,limitedtojustafewcharacters.Thispreventsexistingclassificationschemestobedirectlyappliedinsuchcases,duetosparsenessofthetext.Second,newinformationemergesconstantly,hencebuildingalearningcorpusfrompastpostsmayfailtocapturetheeverevolvinginformationemerginginOSNs.ToovercometheaforementionedlimitationswehavedesignedPythia,anautomatedsystemforshorttextclassificationthatexploitstheWikipediastructureandarticlestoidentifythetopicsoftheposts.Thetopicdiscoveryisperformedintwophases.Inthefirststep,thesystemexploitsWikipediacategoriesandarticlesofthecorrespondingcategoriestobuildthetrainingcorpusforthesuppervisedlearning.Inthesecondstep,thetextofagivenpostisaugmentedusingatextenrichmentmechanismthatextendsthepostwithrelevantWikipediaarticles.Aftertheinitialstepsareperformed,wedeployk-NNclassifiertodeterminethetopic(s)coveredintheoriginalpost.
Data-drivenSerendipityNavigationinUrbanPlaces XiaoyuGe(UniversityofPittsburgh),AmeyaDaphalapurkar(UniversityofPittsburgh),ManaliShimpi(UniversityofPittsburgh),KohliDarpun(UniversityofPittsburgh),KonstantinosPelechrinis(UniversityofPittsburgh),PanosChrysanthis(UniversityofPittsburgh),DemetriosZeinalipour-Yazti(MaxPlanckInstituteforInformaticsandUniversityofCyprus)
WiththeproliferationofmobilecomputingandtheabilitytocollectdetaileddatafortheurbanenvironmentanumberofsystemsthataimatprovidingPointsofInterest(POIs)andtourrecommendationshaveappeared.Theoverwhelmingmajorityofthesesystemsaimsatprovidinganoptimalrecommendation,whereoptimalityreferstoobjectivesofminimizingthedistancetobecoveredormaximizingthequalityofthePOIsrecommended.Amajorproblemisthatbyfocusingontheoptimizationoftheseobjectives,thereremainslittleroomtotheuserforserendipity.Urbanandsocialscientistshaveidentifiedserendipity,i.e.,theabilitytocomeacrossunexpectedplaces,asafeaturethatmakesacitylivable.Inthiswork,weintroduceaprototypeofanexperimentalplatformforevaluatingvenuerecommendationalgorithmsbyprovidinginformativetourrecommendationsbasedonthesuggestedvenues.OurprototypesystemintegratesthenotionofserendipityinurbannavigationatboththevenueaswellastherouterecommendationlevelwithoutcompromisingthequalityanddiversityoftherecommendedPOIs.Inaddition,oursystemallowstheusertouploadtheirownalgorithmsandexploretheirperformanceascomparedtomanywell-knownalgorithms.
TowardAnIntegratedApproachtoLocalizingFailuresinCommunityWaterNetworks(DEMO) QingHan(UniversityofCaliforniaIrvine),PhuNguyen(UniversityofCaliforniaIrvine),RonaldT.Eguchi(ImageCat,Inc.),Kuo-LinHsu(UniversityofCaliforniaIrvine),NaliniVenkatasubramanian(UniversityofCaliforniaIrvine)
Wepresentacyber-physical-human(CPHS)distributedcomputingframework,AquaSCALE,forgathering,analyzingandlocalizinganomalousoperationsofincreasinglyfailure-pronecommunitywaterservices.Today,detectionofwaterpipeleakstakeshourstodays.AquaSCALEleveragesdynamicdatafrommultipleinformationsourcesincludingIoT(InternetofThings)sensingdata,geophysicaldata,humaninputandsimulation/modelingenginestocreateasensor-simulation-dataintegrationplatformthatcanlocatemultiplesimultaneouspipefailuresatfinelevelofgranularitywithhighlevelofaccuracyanddetectiontimereducedbyordersofmagnitude(fromhours/daystominutes).
Demo2:SecurityandPrivacyCluster
PrivateGraph:ACloud-CentricSystemforPrivacy-PreservingSpectralAnalysisofLargeEncryptedGraphs SagarSharma(WrightStateUniversity),KekeChen(WrightStateUniversity)
Graphdatasetshaveinvaluableuseinbusinessapplicationsandscientificresearch.Becauseofthegrowingsizeanddynamicallychangingnatureofgraphs,graphdataownersmaywanttousepubliccloudinfrastructurestostore,process,andperformgraphanalytics.However,whenoutsourcingdataandcomputation,dataownersareatburdentodevelopmethodstopreservedataprivacyanddataownershipfromcuriouscloudproviders.Thisdemonstrationexhibitsaprototypesystemforprivacy-preservingspectralanalysisframeworkforlargegraphsinpublicclouds(PrivateGraph)thatallowsdataownerstocollectgraphdatafromdatacontributors,andstoreandconductsecuregraphspectralanalysisinthecloudwithpreservedprivacyandownership.Thisdemosystemletsitsaudienceinteractivelylearnthemajorcloud-clientinteractionprotocols:theprivacy-preservingdatasubmission,thesecureLanczosandNystromapproximateeigen-decompositionalgorithmsthatworkoverencrypteddata,andtheoutcomeofanimportantapplicationofspectralanalysis-spectralclustering.Intheprocessofdemonstrationtheaudiencewillunderstandtheintrinsicrelationshipamongstcosts,resultquality,privacy,andscalabilityoftheframework.
![Page 109: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/109.jpg)
109
IoTSentinelDemo:AutomatedDevice-TypeIdentificationforSecurityEnforcementinIoT MarkusMiettinen(TechnischeUniversitatDarmstadt),SamuelMarchal(AaltoUniversity),IbbadHafeez(universityofhelsinki),TommasoFrassetto(TechnischeUniversitatDarmstadt),N.Asokan(AaltoUniversity),Ahmad-RezaSadeghi(TechnischeUniversitatDarmstadt),SasuTarkoma(UniversityofHelsinki)
TheemergenceofnumerousnewmanufacturersproducingdevicesfortheInternet-of-Things(IoT)hasgivenrisetonewsecurityconcerns.ManyIoTdevicesexhibitsecurityflawsmakingthemvulnerableforattacksandmanufacturershavedifficultiesinprovidingappropriatesecuritypatchestotheirproductsinatimelyanduser-friendlymanner.Inthispaper,wepresentourimplementationofIOTSENTINEL,whichisasystemaimedatprotectingtheuser’snetworkfromvulnerableIoTdevices.IOTSENTINELautomaticallyidentifiesvulnerabledeviceswhentheyarefirstintroducedtothenetworkandenforcesappropriatetrafficfilteringrulestoprotectotherdevicesfromthethreatsoriginatingfromthevulnerabledevices.
RogueAccessPointDetectorUsingCharacteristicsofChannelOverlappingin802.11nRhonghoJang(INHAuniversityofKorea),JeonilKang(INHAuniversity),AzizMohaisen(SUNYBuffalo),DaehunNyang(DepartmentofComputerandInformationEngineering,INHAUniversity,Incheon,Korea)
Inthiswork,weintroduceapowerfulhardwarebasedrogueaccesspoint(PrAP),whichcanrelaytrafficbetweenalegitimateAPandawirelessstationbackandforth,andactasaman-in-the-middleattacker.OurPrAPisbuiltoftwodedicatedwirelessroutersinterconnectedphysically,andcanrelaytrafficrapidlybetweenastationandalegitimateAP.Throughextensiveexperiments,wedemonstratethatthestateof-the-arttime-basedrogueAP(rAP)detectorscannotdetectourPrAP,althougheffectiveagainstsoftware-basedrAP.TodefendagainstPrAPs,weproposePrAP-Hunterbasedonintentionalchannelinterference.PrAP-Hunterishighlyaccurate,evenunderheavytrafficscenarios.Usingahigh-performance(desktop)andlow-performance(mobile)experimentalsetupsofourPrAPHunterinvariousdeploymentscenarios,wedemonstratecloseto100%ofdetectionrate,comparedto60%detectionratebythestate-of-the-art.WeshowthatPrAP-Hunterisfast(takes5-10sec),doesnotrequireanypriorknowledge,andcanbedeployedinthewildbyrealworldexperimentsat10coffeeshops.Keywords.Intrusiondetection,WirelessLAN,RogueAP,channelinterference,IEEE802.11n.
ReverseCloak:AReversibleMulti-levelLocationPrivacyProtectionSystem ChaoLi(UniversityofPittsburgh),BalajiPalanisamy(UniversityofPittsburgh),AravindKalaivanan(UniversityofPittsburgh),SriramRaghunathan(UniversityofPittsburgh)
Withthefastpopularizationofmobiledevicesandwirelessnetworks,alongwithadvancesinsensingandpositioningtechnology,wearewitnessingahugeproliferationofLocation-basedServices(LBSs).LocationanonymizationreferstotheprocessofperturbingtheexactlocationofLBSusersasacloakingregionsuchthatauser'slocationbecomesindistinguishablefromthelocationofasetofotherusers.However,existinglocationanonymizationtechniquesfocusprimarilyonsinglelevelunidirectionalanonymization,whichfailstocontroltheaccesstothecloakingdatatoletdatarequesterswithdifferentprivilegesgetinformationwithvaryingdegreesofanonymity.Inthisdemonstration,wepresentatoolkitforReverseCloak,alocationperturbationsystemtoprotectlocationprivacyoverroadnetworksinamulti-levelreversiblemanner,consistingofan'Anonymizer'GUItoadjusttheanonymizationsettingsandvisualizethemultilevelcloakingregionsoverroadnetworkforlocationdataownersanda'De-anonymizer'GUItode-anonymizethecloakingregionanddisplaythereducedregionoverroadnetworkforlocationdatarequesters.Withthetoolkit,wedemonstratethepracticalityandeffectivenessoftheReverseCloakapproach.
Demo3:CloudsandVirtualizationCluster
Hopsworks:ImprovingUserExperienceandDevelopmentonHadoopwithScalable,StronglyConsistentMetadata MahmoudIsmail(KTH-RoyalInstituteofTechnology),ErmiasGebremeskel(RISESICS),TheofilosKakantousis(RISESICS),GautierBerthou(RISESICS),JimDowling(KTH-RoyalInstituteofTechnology)
Hadoopisapopularsystemforstoring,managing,andprocessinglargevolumesofdata,butithasbare-bonesinternalsupportformetadata,asmetadataisabottleneckandlessmeansmorescalability.Theresultisascalableplaformwithrudimentaryaccesscontrolthatisneitheruser-nordeveloper-friendly.Also,metadataservicesthatarebuiltonHadoop,suchasSQL-on-Hadoop,accesscontrol,dataprovenance,anddatagovernancearenecessarilyimplementedaseventuallyconsistentservices,resultinginincreaseddevelopmenteffortandmorebrittlesoftware.Inthispaper,wepresentanewproject-basedmulti-tenancymodelforHadoop,builtonanewdistributionofHadoopthatprovidesadistributeddatabasebackendfortheHadoopDistributedFilesystem’s(HDFS)metadatalayer.WeextendHadoop’smetadatamodeltointroduceprojects,datasets,andproject-usersasnewcoreconceptsthatenableauser-friendly,UI-drivenHadoopexperience.Asourmetadataserviceisbackedbyatransactionaldatabase,developerscaneasilyextendmetadatabyaddingnewtablesandensurethestrongconsistencyofextendedmetadatausingbothtransactionsandforeignkeys.
IsolationinDockerthroughLayerEncryption IoannisGiannakopoulos(NationalTechnicalUniversityofAthens),KonstantinosPapazafeiropoulos(NationalTechnicalUniversityofAthens),KaterinaDoka(NationalTechnicalUniversityofAthens),NectariosKoziris(NationalTechnicalUniversityofAthens)
Containersareconstantlygaininggroundinthevirtualizationlandscapeasalightweightandefficientalternativetohypervisor-basedVirtualMachines,withDockerbeingthemostsuccessfulrepresentative.Dockerreliesonunion-capablefilesystems,whereanyactionperformedtoabaseimageiscapturedasanewfilesystemlayer.ThisstrategyallowsdeveloperstoeasilypackapplicationsintoDockerimagelayersanddistributethemviapublicregistries.However,thisimagecreationanddistributionstrategydoesnotprotectsensitivedatafrommaliciousprivilegedusers(e.g.,registryadministrator,cloudprovider),sinceencryptionisnotnativelysupported.WeproposeanddemonstrateamechanismforsecureDockerimagemanipulationthroughoutitslifecycle:Thecreation,storageandusageofaDockerimageisbackedbyadata-at-restmechanism,whichmaintainssensitivedataencryptedondiskandencrypts/decryptsthemon-the-flyinordertopreservetheirconfidentialityatalltimes,whilethedistributionandmigrationofimagesisenhancedwithamechanismthatencryptsonlyspecificlayersofthefilesystemthatneedtoremainconfidentialandensuresthatonlylegitimatekeyholderscandecryptthemandreconstructtheoriginalimage.Througharichinteractionwithoursystemtheaudiencewillexperiencefirst-handhowsensitiveimagedatacanbesafelydistributedandremainencryptedatthestoragedevicethroughoutthecontainer'slifetime,bearingonlyamarginalperformanceoverhead.
Dela-SharingLargeDatasetsbetweenHadoopClusters AlexandruA.Ormenisan(KTHRoyalInstituteofTechnology),JimDowling(KTHRoyalInstituteofTechnology)
![Page 110: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/110.jpg)
110
Bigdatahas,inrecentyears,revolutionisedanever-growingnumberoffields,frommachinelearningtoclimatesciencetogenomics.Thecurrentstate-of-the-artforstoringlargedatasetsiseitherobjectstoresordistributedfilesystems,withHadoopbeingthedominantopen-sourceplatformformanaging‘BigData’.Existinglarge-scalestorageplatforms,however,lacksupportfortheefficientsharingoflargedatasetsovertheInternet.Thosesystemsthatarewidelyusedforthedisseminationoflargefiles,likeBitTorrent,needtobeadaptedtohandlechallengessuchasnetworklinkswithbothhighlatencyandhighbandwidth,andscalablestoragebackendsthatareoptimisedforstreamingandnotrandomaccess.Inthispaper,weintroduceDela,apeer-to-peerdata-sharingserviceintegratedintotheHopsHadoopplatformthatprovidesanend-to-endsolutionfordatasetsharing.Delaisdesignedforlarge-scalestoragebackendsanddatatransfersthatarebothnon-intrusivetoexistingTCPnetworktrafficandprovidehighernetworkthroughputthanTCPonhighlatency,highbandwidthnetworklinks,suchastransatlanticnetworklinks.Delaprovidesapluggablestoragelayer,implementingtwoalternativewaysforclientstoaccessshareddata:streamprocessingofdataasitarriveswithKafka,andtraditionalofflineaccesstodatausingtheHadoopDistributedFilesystem.DelaisthefirststepfortheHadoopplatformtowardscreatinganopendatasetecosystemthatsupportsuser-friendlypublishing,searching,anddownloadingoflargedatasets.
InVivoEvaluationoftheSecureOpportunisticSchemesMiddlewareusingaDelayTolerantSocialNetworkCoreyE.Baker(UniversityofCaliforniaSanDiego),AllenStarke(UniversityofFlorida),TanishaG.Hill-Jarrett(UniversityofFlorida),JaniseMcNair(UniversityofFlorida)
Overthepastdecade,onlinesocialnetworks(OSNs)suchasTwitterandFacebookhavethrivedandexperiencedrapidgrowthtoover1billionusers.AmajorevolutionwouldbetoleveragethecharacteristicsofOSNstoevaluatetheeffectivenessofthemanyroutingschemesdevelopedbytheresearchcommunityinreal-worldscenarios.Inthisdemonstration,weshowcasetheSecureOpportunisticSchemes(SOS)middlewarewhichallowsdifferentroutingschemestobeeasilyimplementedrelievingtheburdenofsecurityandconnectionestablishment.ThefeasibilityofcreatingadelaytolerantsocialnetworkisdemonstratedbyusingSOStoenableAlleyOopSocial,asecuredelaytolerantnetworkingresearchplatformthatservesasareal-lifemobilesocialnetworkingapplicationforiOSdevices.AlleyOopSocialallowsuserstointeract,publishmessages,anddiscoverothersthatsharecommoninterestsinanintermittentnetworkusingBluetooth,peer-to-peerWiFi,andinfrastructureWiFi.
Demo4:DistributedSystemsandNetworkingCluster
ScalingandLoadTestingLocation-basedPublishandSubscribe BertilChapuis(UniversityofLausanne),BenoîtGarbinato(UniversityofLausanne)
TheriseoftheInternetofthings(IoT)posesmassivescalabilityissuesforlocation-basedservices.Moreparticularly,location-awarepublishandsubscribeservicesarestrugglingtoscaleoutthecomputationofmatchesbetweenpublicationsandsubscriptionsthatcontinuouslyupdatetheirlocation.Inthisdemonstrationpaper,weproposeanoveldistributedandhorizontallyscalablearchitectureforlocation-awarepublishandsubscribe.Ourmiddlewarearchitecturereliesonamultisteproutingmechanismbasedonconsistenthashingandrangepartitioning.Todemonstrateitsscalability,wepresentatrafficdatagenerator,which,incontrasttoexistinggenerators,canbeusedtoperformreal-timeloadtests.Finally,weshowthatourarchitecturecanbedeployedonasmall10-nodeclusterandcanprocessupto80,000locationupdatespersecondproducing25,000matchesperseconds.
ADistributedEvent-centricCollaborativeWorkflowsDevelopmentSystemforIoTApplication YongyangCheng(BUPT),ShuaiZhao(BUPT),BoCheng(BUPT),ShouluHou(BUPT),XiuleiZhang(BUPT),JunliangChen(BUPT)
TherapiddevelopmentofInternetofThings(IoT)attractsgrowingattentionfrombothindustryandacademia.IoTseamlesslyconnectsthephysicalworldandcyberspaceviavarioussensors.Itismoreworthforustopayattentiontothemechanismoftheeventstoworkcollaborativelyratherthanthosestandalonesensors.Inthispaper,wepresentaDistributedEvent-centricCollaborativeWorkflowsdevelopmentsystemforIoTapplication,calledDECW.Itsupportslooselycoupledevent-basedinteractionbetweenprocesses,whichenablesreal-timeresponsetoeventsfromthephysicalworld.Unliketraditionalcentralizedcontrolflowmode,theinteractionbetweenprocessesinDECWisconstrainedbytheeventinterface.Userscoulddynamicallyadjusttheinterfacebetweenprocesseswithoutmodifyingtheinternallogicoftheprocess.Inaddition,DECWsystemprovidesafulllifecycleforthedevelopmentandoperationoftheIoTapplication,includinggraphicalcreationofprocesses,dynamicdefinitionoftheprocessinteractioninterfaces,logicalvalidation,distributedpackaginganddeployment,parallelexecution,andreal-timemonitoringandmanagingtherunningstatusoftheIoTapplication.
IncentiveMechanismforData-CentricMessageDeliveryinDelayTolerantNetworks HimanshuJethawa(MissouriUniversityofScienceandTechnology),SanjayMadria(MissouriUniversityofScienceandTechnology)
Akeyissueindelaytolerantnetworks(DTN)istofindtherightnodetostoreandrelaymessages.Weconsidermessagesannotatedwiththeuniquekeywordsdescribingthemessagesubject,andnodesalsoaddskeywordstodescribetheirmissioninterests,priorityandtheirtransientsocialrelationship(TSR).Tooffsetresourcecosts,anincentivemechanismisdevelopedovertransientsocialrelationshipswhichenrichenroutemessagecontentandmotivatebettersemanticallyrelatednodestocarryandforwardmessages.Theincentivemechanismensuresavoidanceofcongestionduetouncooperativeorselfishbehaviorofnodes.
PerformanceOfCognitiveWirelessChargerForNear-FieldWirelessChargingSang-YoonChang(UniversityofColorado,ColoradoSpringsandAdvancedDigitalSciencesCenter),SristiLakshmiSravanaKumar(AdvancedDigitalSciencesCenter),Yih-ChunHu(UniversityofIllinoisatUrbana-Champaign)
WirelesschargingprovidesaconvenientwaytochargevariousmobileandIoTdevices.State-of-the-artwire-lesschargingsystemsoperateataparticularfrequencyandarecontrolledbyexplicitnetworkingwiththepower-receivingdevices.However,controlbyexplicitnetworkingisnotdesignedtocopewiththevariationsinthepower-receivingdevicesplacementsandalignments.Thisisespeciallymoresignificantinnear-fieldandpseudo-tightlycoupledchargingapplicationsasmorechargingpadsarebeingdeployedinthepublicdomainsandservingheterogeneousclients.Weestablishthatfrequencycontrolachievesbetterperformancegainsininductive-couplingchargingapplicationsandisalsosensitivetothevariationsintheplacementandalignmentbetweenthepower-transmittingandthepower-receivingcoils.Inthisdemo,weshowtheimpactinpowertransferperformancecausedbythevariationsintheplacementandalignmentbetweenthepower-transmittingandthepower-receivingcoilsandshowcasetheperformanceofourcognitivewirelesscharger(CWC),whichadaptivelycontrolstheoperatingfrequencyinreal-timeusingimplicitfeedbackforoptimaloperations.
![Page 111: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/111.jpg)
111
![Page 112: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/112.jpg)
112
Poster Track Paper Abstracts
Poster1:DistributedApplicationsCluster
TowardVehicleSensing:Anintegratedapplicationwithsparsevideocamerasandintelligenttaxicabs YangWang(UniversityofScienceandTechnologyofChina),WujiChen(UniversityofScienceandTechnologyofChina),WeiZheng(Sanofi-AventisUSLLC),HeHuang(SoochowUniversity),WenZhang(UniversityofScienceandTechnologyofChina),HengchangLiu(UniversityofScienceandTechnologyofChina)
Duetothesparsedistributionofroadvideosurveillancecameras,precisetrajectorytrackingforvehiclesremainsachallengingtask.Tothebestofourknowledge,noneofthepreviousresearchconsideredusingon-roadtaxicabsasmobilevideosurveillancecamerasandroadtrafficflowpatterns,thereforenotsuitableforrecoveringtrajectoriesofvehicles.Withthisinsight,wemodelthetraveltime-costofaroadsegmentduringvarioustimeperiodspreciselywithLNDs(LogarithmicNormalDistributions),thenuseLSNDs(LogSkewNormalDistributions)toapproximatethetime-costofanurbantripduringvarioustimeperiods.Weproposeanapproachtocalculatepossiblelocationandtimedistributionofthevehicle,selectthetaxicabtoverifythedistributionbyuploadingandcheckingvideoclipsofthistaxicab,finallyrefinetherestoringtrajectoryinarecursivemanner.Weevaluateoursolutiononreal-worldtaxicabandroadsurveillancesystemdatasets.Experimentalresultsdemonstratethatourapproachoutperformsalternativesolutionsintermsofaccuracyratioofvehicletracking.
SegmentationofTimeSeriesbasedonKineticCharacteristicsforStorageConsumptionPrediction BeibeiMiao(Baidu,Inc),YuChen(Baidu,Inc),XueboJin(SchoolofComputerandInformationEngineering,BeijingTechnologyandBusinessUniversity),BoWang(Baidu,Inc),XianpingQu(Baidu,Inc),DongWang(Baidu,Inc),ShiminTao(Baidu,Inc),ZhiZang(Baidu,Inc)
TheInternetservicesgeneratehugeamountofdata,whichrequirelargespaceforstorage.Determiningdevicepurchaseplanturnsouttobeveryimportantfortheserviceproviders.Under-purchasingmightleadtodataloss,whileover-purchasingwouldresultinwaste.Inthispaper,weproposealinearregressionbasedapproachtopredictthestoragedemandaccordingtothetimeseriesofthestorageconsumption.Wepartitionedthestorageconsumptiontimeseriesintoseverallinearsegments,andperformpredictiononthelastsegmentusinglinearregression.Sincethepositionofturningpointsbetweenadjacentsegmentsandthetotalnumberofthesegmentsarebothunknown,howtoachievetheonlinesegmentationbecomesabigchallenge.Aimingtosolvethisproblem,wecarriedouttheKalmanAnovasegmentationmethod.Experimentresultsshowthatourmethodhasgoodaccuracyinprecision,recallandF-measurevalues.Moreover,themethodisabletosegmentnonlineartimeseriesaswell,suggestingapotentialwiderapplication.TheproposedmethodhasbeendeployedinBaiduInc.andsavesabout45thousanddollarsinoneofitsdevicepurchaseprogram.
AMulti-stageHierarchicalWindowModelwithApplicationtoReal-TimeGraphAnalysis SachiniJayasekara(UniversityofMelbourne),ShanikaKarunasekera(UniversityofMelbourne),AaronHarwood(UniversityofMelbourne)
Thedynamicnatureofreal-worldnetworks,suchassocialnetworksandcommunicationnetworks,hasincreasedthefocustowardsreal-timedynamicgraphanalysis.Observationsmadebasedonreal-timeanalysisofdynamicgraphsreflectthelatestpropertiesofthegraphandhavethemostvalueinreal-timeanalysis.Computinggraphpropertiesoflarge-scale,fast-evolvinggraphsinreal-timeischallengingdue,notonlytothehighcomputationalandmemorycost,butalsototheunderstandingoftheresultwithrespecttothedatafromwhichitwasderived.Thispaperproposesamulti-stagehierarchicalwindowmodelthatcanaidinrigorousunderstandingofcomplicatedreal-timeresultsandweapplyittogenerategraphsbasedonreal-timeupdatesalongwithperiodiccomputationsongraphsnapshotsforprocessingdynamicgraphs.Moreover,thepaperdiscussestheutilizationofparallelwindowcomputation.ThepaperevaluatesthehierarchicalmodelthroughanalyzinggraphsformedbycooccurringhashtagsinaTwitterdata-stream.
DynamicPricingatElectricVehicleChargingStationsforQueueingDelayReduction XiaoshanSun(UniversityofScienceandTechnologyofChina),PengXu(UniversityofScienceandTechnologyofChina),JinyangLi(UniversityofScienceandTechnologyofChina),HengchangLiu(UniversityofScienceandTechnologyofChina),WeiZheng(Sanofi-Aventis)
Theresearchofelectricvehicles(EVs)hasgainedmoreandmoreattentioninrecentyearsinbothindustryandacademia,andnewregistrationsofEVsincreaserapidly,whilethelongdelayattheconvenientbutcrowdedchargingstationsmaydiscouragemanydriversfromswitchingtoEVs.Toaddresstheproblems,weproposeanoveldynamicpricingpolicythatallowschargingstationstoadjusttheirservicefeesinrealtimebasedontheloadatthestations.Inourwork,theselectionofdriversismodeledbyanewdissatisfactionfunctionwithmultiplevariables,whichcanbeeasilyvalidatedandimprovedbyrealapplications,andoursolutionisevaluatedfromtherealworlde-chargedataset.Tothebestofourknowledge,thisisthefirstworkthatconsidersdynamicservicefeesamongvariouschargingstationsforloadbalancingandreductionofqueueingdelay.Thismakesourworkmorerealisticandbeneficial.
PairwiseRankingAggregationbyNon-interactiveCrowdsourcingwithBudgetConstraints ChangjiangCai(StevensInstituteofTechnology),HaipeiSun(StevensInstituteofTechnology),BoxiangDong(MontclairStateUniversity),BoZhang(StevensInstituteofTechnology),TingWang(LehighUniversity),WendyHuiWang(StevensInstituteofTechnology)
Crowdsourcedrankingalgorithmsaskthecrowdtocomparetheobjectsandinferthefullrankingbasedonthecrowdsourcedpairwisecomparisonresults.Inthispaper,weconsiderthesettinginwhichthetaskrequesterisequippedwithalimitedbudgetthatcanaffordonlyasmallnumberofpairwisecomparisons.Tomaketheproblemmorecomplicated,thecrowdmayreturnnoisycomparisonanswers.Weproposeanapproachtoobtainagood-qualityfullrankingfromasmallnumberofpairwisepreferencesintwosteps,namelytaskassignmentandresultinference.Inthetaskassignmentstep,wegeneratepairwisecomparisontasksthatproduceafullrankingwithhighprobability.Intheresultinferencestep,basedonthetransitivepropertyofpairwisecomparisonsandtruthdiscovery,wedesignanefficientheuristicalgorithmtofindthebestfullrankingfromthepotentiallyconflictivepairwisepreferences.Theexperimentresultsdemonstratetheeffectivenessandefficiencyofourapproach.
Buffer-BasedReinforcementLearningforAdaptiveStreaming YueZhang(SUNYBinghamton),YaoLiu(SUNYBinghamton)
![Page 113: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/113.jpg)
113
Adaptivestreamingimprovesuser-perceivedqualitybyalteringthestreamingbitratedependingonnetworkconditions,tradingreducedvideobitratesforreducedstalltimes.Existingadaptationapproaches,e.g.,rate-based,bufferbased,eitherrelyheavilyonaccuratebandwidthpredictionorcanbeoverly-conservativeaboutvideobitrates.Inthiswork,weproposeareinforcementlearningapproachtochoosethesegmentqualityduringplayback.Thisapproachusesonlythebufferstateinformationandoptimizesforameasureofuser-perceivedstreamingquality.SimulationresultsshowthatourproposedapproachachievesbetterQoEthanrate-,buffer-basedapproaches,aswellasotherreinforcementlearningapproaches.
Thecaseforusingcontent-centricnetworkingfordistributinghigh-energyphysicssoftware MohammadAlhowaidi(UniversityofNebraska-Lincoln),ByravRamamurthy(UniversityofNebraska-Lincoln),BrianBockelman(UniversityofNebraska-Lincoln),DavidSwanson(UniversityofNebraska-Lincoln)
NamedDataNetworking(NDN)isoneofthepromisingfutureinternetarchitectures,whichfocusesonthedataratherthanitslocation(IP/host-basedsystem).NDNhasseveralcharacteristicswhichfacilitateaddressingandroutingthedata:fail-over,in-networkcachingandloadbalancing.Thismakesitusefulinareassuchasmanagingscientificdata.TheCMSexperimentontheLargeHadronCollider(LHC)hasadataaccessproblemamenabletocontent-centricnetworking.CERNVirtualMachineFileSystem(CVMFS)isusedbyHighEnergyPhysics(HEP)communityforworldwidesoftwaredistribution.CVMFSmaintainitsdatabyusingcontentaddressablestorage,whichmakesitsuitableforNDN.Inthispaper,weinvestigatethepossibilitiesofusingacontentcentricnetworkingarchitecturesuchasNDNondistributingCMSsoftware.
LAVEA:Latency-awareVideoAnalyticsonEdgeComputingPlatformShanheYi(CollegeofWilliamandMary),ZijiangHao(CollegeofWilliamandMary),QingyangZhang(WayneStateUniversity),QuanZhang(WayneStateUniversity),WeisongShi(WayneStateUniversity),QunLi(CollegeofWilliamandMary)
WepresentLAVEA,asystembuiltforedgecomputing,whichoffloadscomputationtasksbetweenclientsandedgenodes,collaboratesnearbyedgenodes,toprovidelow-latencyvideoanalyticsatplacesclosertotheusers.Wehaveutilizedanedge-firstdesigntominimizetheresponsetime,andcomparedvarioustaskplacementschemestailedforinter-edgecollaboration.Ourresultsrevealthattheclient-edgeconfigurationhastaskspeedupagainstlocalorclient-cloudconfigurations.
CompleteToleranceRelationbasedFillingAlgorithmusingSpark JinglingYuan(WuhanUniversityofTechnology),YaoXiang(WuhanUniversityofTechnology),XianZhong(WuhanUniversityofTechnology),MinchengChen(WuhanUniversityofTechnology),TaoLi(UniversityofFlorida)
Withtheadventofcloudcomputing,renewableenergyisintegratedintodatacenterpowersupplysystemsincreasingly.Thepowerstatisticscollectionmaynotbeavailableduetotheinstabilityofrenewableenergy,whichresultsinincompletedata.Theincompleteenergydatawillsignificantlydisturbthemanagementofdatacenters.Wefurtherproposeafillingalgorithmbasedoncompletetoleranceclass.Thealgorithmexpandsthetraditionaltolerancerelation,andfillsthemissingvaluesoftheenergydata,whichensuresthedataintegrity.Bytakinggoodadvantageofin-MemoryComputing,WefurtherparallelizeandoptimizeouralgorithmusingSpark.Theexperimentresultsdemonstratethatouralgorithmoutperformsothergeneralfillingalgorithmsintermsoffillingaccuracy.Theproposedalgorithmalsoshowsgoodperformanceasthemissingraterisesup.
Poster2:SecurityandPrivacyCluster
TowardsSecurePublicDirectoryforPrivacy-PreservingDataSharingAminFallahi(SyracuseUniversity),XiLiu(SyracuseUniversity),YuzheTang(SyracuseUniversity),ShuangWang(UCSD),RuiZhang(ChineseAcademyofSciences)
Inemergingfederateddatabasesystems,suchasHealthInformationExchange(orHIE),animportantyetunderstudiedproblemistheprivacy-preservingsharingofpersonalrecordsamongautonomousdataowners.Thegoalposestechnicaldesignchallenges,includingtheassuredprivacypreservationunderbackground-knowledgeattacks,andscalableandsecuremulti-partycomputationsonprivatebig-datainalarge-scalesystem.Totacklethechallenges,weproposeaprotocol,multi-partydeterministicnoisingorMPDN,whichdeterministicallyinjectsnoisestothepublishedmeta-datawhilestayingawareofthebackgroundknowledge.Italsooptimizestheperformanceofmulti-partycomputation(orMPC)bypre-computationonthepublicbackgroundknowledge.Thepre-computationexhibitsdata-levelparallelismandweleveragegeneral-purposecomputingongraphicsprocessingunits(GPGPU)inourimplementationtoexploittheparallelismandtofurtheroptimizeperformance.Theproposedprotocolisimplementedonopen-sourceMPCsoftware(i.e.,GMW)anditsefficiencywithaspeedupofmorethananorderofmagnitudeisdemonstratedinageo-distributedsetting.Throughevaluationonreal-worlddatasets,theassuranceofprivacypreservationisalsoverified.
AnonymousRoutingtoMaximizeDeliveryRatesinDTNs KazuyaSakai(TokyoMetropolitanUniversity),Min-TeSun(NationalCentralUniversity),Wei-ShinnKu(AuburnUniversity),JieWu(TempleUniversity)
Securityandprivacyissuesareconsideredtobetwoofthemostsignificantconcernstoorganizationsandindividualsusingmobileapplications.Inthispaper,weseektoaddressanonymouscommunicationsindelaytolerantnetworks(DTNs).Whilemanydifferentanonymousroutingprotocolshavebeenproposedforadhocnetworks,tothebestofourknowledge,onlyvariantsofonion-basedroutinghavebeentailoredforDTNs.Sinceeachtypeofanonymousroutingprotocolhasitsadvantagesanddrawbacks,thereisnosingleanonymousroutingprotocolforDTNsthatcanadapttothedifferentlevelsofsecurityrequirements.Inthispaper,wefirstdesignasetofanonymousroutingprotocolsforDTNs,calledanonymousEpidemicandzone-basedanonymousrouting,basedontheoriginalanonymousroutingprotocolsforadhocnetworks.Then,weproposeaframeworkofanonymousrouting(FAR)forDTNs,whichsubsumesalltheaforementionedprotocols.Bytuningitsparameters,theproposedFARisabletooutperformonionbased,anonymousEpidemic,andzone-basedrouting.Inaddition,numericalanalysesforthetraceablerateandnodeanonymitymodelsarebuilt.Extensivesimulationsusingrandomlygeneratedgraphsaswellasrealtracesareconductedtodemonstratethatgivenappropriateparametersettings,ourFARoutperformsalltheexistinganonymousroutingprotocolsforDTNs.
EvaluatingConnectionResiliencefortheOverlayNetworkKademliaHennerHeck(UniversitätKassel),OlgaKieselmann(UniversitätKassel),ArnoWacker(UniversitätKassel)
![Page 114: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/114.jpg)
114
Kademliaisadecentralizedoverlaynetwork,uptonowmainlyusedforhighlyscalablefilesharingapplications.Duetoitsdistributednature,itisfreefromsinglepointsoffailure.Communicationcanhappenoverredundantnetworkpaths,whichmakesinformationdistributionwithKademliaresilientagainstfailingnodesandattacks.Inthispaper,wesimulateKademlianetworkswithvaryingparametersandanalyzethenumberofnode-disjointpaths.Withourresults,weshowtheinfluenceoftheseparametersonthenetworkconnectivityand,therefore,theresilienceagainstfailingnodesandcommunicationchannels.
Shortfall-basedOptimalSecurityProvisioningforInternetofThings AntoninoRullo(UniversityofCalabria),EdoardoSerra(BoiseStateUniversity),JorgeLobo(UniversitatPompeaFabra),ElisaBertino(PurdueUniversity)
WepresentaformalmethodforcomputingthebestsecurityprovisioningforInternetofThings(IoT)scenarioscharacterizedbyahighdegreeofmobility.Thesecurityinfrastructureisintendedasasecurityresourceallocationplan,computedasthesolutionofanoptimizationproblemthatminimizestheriskofhavingIoTdevicesnotmonitoredbyanyresource.Weemploytheshortfallasariskmeasure,aconceptmostlyusedintheeconomics,andadaptittoourscenario.Weshowhowtocomputeandevaluateanallocationplan,andhowsuchsecuritysolutionsaddressthecontinuoustopologychangesthataffectanIoTenvironment.
GroupDifferentialPrivacy-preservingDisclosureofMulti-levelAssociationGraphs BalajiPalanisamy(UniversityofPittsburgh),ChaoLi(UniversityofPittsburgh),PrashantKrishnamurthy(UniversityofPittsburgh)
Traditionalprivacy-preservingdatadisclosuresolutionshavefocusedonprotectingtheprivacyofindividual’sinformationwiththeassumptionthatallaggregate(statistical)informationaboutindividualsissafefordisclosure.Suchschemesfailtosupportgroupprivacywhereaggregateinformationaboutagroupofindividualsmayalsobesensitiveandusersofthepublisheddatamayhavedifferentlevelsofaccessprivilegesentitledtothem.WeproposethenotionofGroupDifferentialPrivacythatprotectssensitiveinformationofgroupsofindividualsatvariousdefinedprivacylevels,enablingdatauserstoobtainthelevelofaccessentitledtothem.Wepresentapreliminaryevaluationoftheproposednotionofgroupprivacythroughexperimentsonrealassociationgraphdatathatdemonstratetheguaranteesongroupprivacyonthediscloseddata.
TrackingInformationFlowinCyber-PhysicalSystems StefanGries(UniversityofDuisburg-Essen),MarcHesenius(UniversityofDuisburg-Essen),VolkerGruhn(UniversityofDuisburg-Essen)
Cyber-PhysicalSystemsaredistributed,heterogeneous,decentralizedandlooselycouplednetworksinwhichindividualsystemsmeasurephysicalprocesses,exchangeinformation,andinfluenceprocesses.Sensorsmeasurethesephysicalprocesses,whileaggregatorsprocessthemandactuatorsperformresultingactions.Decisionsareoftenbasedonsensordatacollectedbyothersystems.Furthermore,theaggregatorsalsointerchangeinformationandusethemtoderiveowndecisions.Decisionsmustbecomprehensible.However,thisisonlythecaseifalldatadependenciesareknown.Duetothesizeofthesenetworks,theirloosecouplingandtheirdynamicbehavior,decisionsmadebyasystemarenotalwayseasytounderstand.Ifanerroroccursinthesystem,theerrorsourcemustbeidentified.Itmustbeknownonwhichdataadecisionwasbased.However,sincethedecisioncanbebasedoninformationfromothernodes,thesearchfortheerrorsourceisnotatrivialtask.Keepinmind,thatdependentnodescanhavedependenciesthemselvesaswell.WepresenttheInformationFlowMonitor(IFM)thatcollectsinformationaboutsemanticdatadependenciesindynamicnetworks.Thecollecteddependencyinformationisprovidedatacentralnetworklocation.Subsequently,semanticdependenciesbetweeninformationcanbevisualized.
Privacy-preservingMatchmakinginGeosocialNetworkswithUntrustedServers QiuxiangDong(ArizonaStateUniversity),DijiangHuang(ArizonaStateUniversity)
AsamajorbranchofLBSs,geosocialnetworkingservicesbecomepopular.Animportantfunctionalityofgeosocialnetworkingservicesisallowingpeopletofindpotentialfriendswhohavesimilarprofilewithincloseproximityandinitiatecommunicationwitheachother.However,inordertorealizethisfunctionality,mostexistingservicesrequiremobileuserstorevealtheirprofilesandlocationinformationtoanuntrustedserviceprovider,whichmayexposeLBSstovulnerabilitiesforabuseandendangermobileusers’privacy.Toaddressthisproblem,weproposetoencryptusers’profilewithanewsearchableencryptionscheme.Combiningthissearchableencryptionschemewithothercryptographictechniquesweconstructaprivacypreservingmatchmakingsystem.Comparedwithapreviousonethataimstosolvethesameproblem,oursismoresecure,supportsmoreflexiblefunctionalitiesandmovescomputationallyheavykeyupdatestoresourcefulserviceproviders.
You’veBeenTricked!AUserStudyoftheEffectivenessofTyposquattingTechniques JeffreySpaulding(SUNYBuffalo),ShambhuUpadhyaya(SUNYBuffalo),AzizMohaisen(SUNYBuffalo)
ThenefariouspracticeofTyposquattinginvolvesdeliberatelyregisteringInternetdomainnamescontainingtypographicalerrorsthatprimarilytargetpopulardomainnames,inanefforttoredirectuserstounintendeddestinationsorstealingtrafficformonetarygain.Typosquattinghasexistedforwellovertwodecadesandcontinuestobeacrediblethreattothisday.AsrecentlyshownintheonlinemagazineSlate.com[16],cybercriminalshaveattemptedtodistributemalwarethroughNetflix.om,atyposquattedvariantofthepopularstreamingsiteNetflix.comthatusesthecountrycodetop-leveldomain(ccTLD)forOman(.om).Whilemuchofthepriorworkhasexaminedvarioustyposquattingtechniquesandhowtheychangeovertime,nonehaveconsideredhoweffectivetheyareindeceivingusers.Inthispaper,weattempttofillinthisgapbyconductingauserstudythatexposessubjectstoseveraluniformresourcelocators(URLs)inanattempttodeterminetheeffectivenessofseveraltyposquattingtechniquesthatareprevalentinthewild.WealsoattempttodetermineifthesecurityeducationandawarenessofcybercrimessuchastyposquattingwillaffectthebehaviorofInternetusers.Ultimately,wefoundthatsubjectstendtocorrectlyidentifytyposquattingwhichaddscharacterstothedomainnames,whilethemosteffectivetechniquestodeceiveusersinvolvespermutationsandsubstitutionsofcharacters.Wealsofoundthatsubjectsgenerallyperformedbetterandfasteratidentifyingtyposquatteddomainnamesafterbeingthoroughlyeducatedaboutthem,andthatcertainattributessuchasAgeandEducationaffecttheirbehaviorwhenexposedtothem.
Real-timeDetectionofIllegalFileTransfersintheCloud AdamBowers(MissouriUniversityofScienceandTechnology),DanLin(MissouriUniversityofScienceandTechnology),AnnaSquicciarini(ThePennsylvaniaStateUniversity),AliHurson(MissouriUniversityofScienceandTechnology)
Therehasbeenaprolificriseinthepopularityofcloudstorageinrecentyears.Whilecloudstorageoffersmanyadvantagessuchasflexibilityandconvenience,usersarenowunabletotellorcontroltheactuallocationsoftheirdata.Thislimitationmayaffectusers’confidenceandtrustinthestorageprovider,orevenbeunsuitableforstoringdatawithstrictlocationrequirements.Toaddressthisissue,weproposeanillegalfiletransferdetectionframeworkthatconstantlymonitors
![Page 115: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/115.jpg)
115
thereal-timefiletransfersinthecloudandiscapableofdetectingpotentialillegaltransferswhichmovessensitivedataoutsidethe(“legal”)boundariesspecifiedbythefileowner.Themainideaistoclassifyingmultipleusers’locationpreferenceswhenmakingthedatastoragearrangementinthecloudnodes.Wemodelthelegalfiletransfersamongnodesasaweightedgraphandthenmaximizetheprobabilityofstoringdataitemsofsimilarprivacypreferencesinthesameregion.ThenweleveragethesocketmonitoringfunctionsprovidedbyLAST-HDFS(arecentlocation-awareHadoopfilestoragesystem)tomonitorthereal-timecommunicationamongcloudnodes.Basedonourlegalfiletransfergraphandthedetectedcommunication,weproposeanapproachtocalculatetheprobabilityofthedetectedtransfertobeillegal.Wehaveimplementedourproposedframeworkandourexperimentalresultsindicatethatourapproachisabletodetectmuchmoreillegalfiletransfersthanthestateoftheart.
EyesoftheSwarm:Streamers'DetectioninBitTorrent DanielSilva(FluminenseFederalUniversity),AntonioRocha(FluminenseFederalUniversity)
ManyBitTorrent(BT)clientsareusingtheseBTnetworksasavideo-on-demandservice,takingadvantageofthepopularityandthelargecollectionofmediaavailable.However,transformingtheswarmsintoanon-demandmediaservicecancauseseriousdamagetotheoverallnetworkperformance.Inthispaper,weproposeamethodology,usingtheconceptsofEntropy,andpresentaSpyBitTorrentclientthatisabletoidentifypeersstreaminginaswarm.Largescalemonitoring,forrealswarms,wereperformedtodetectthepresenceofstreamers.
Poster3:CloudsandVirtualizationCluster
Loadpredictionforenergy-awareschedulingforCloudscomputingplatformsAlexandreDambreville(LRI),JoannaTomasik(CentraleSupélec),JohanneCohen(LRI-CNRS),FabienDufoulon(LRI)
WeaddressonlineschedulingforserversofCloudserviceproviders.Eachserveriscomposedofseveralvariablespeedprocessorswhosepowerfunctionisconvex.Theserversmaybebusy,idleorswitchedoff.TheobjectiveofourschedulingistominimizetheenergyconsumedbyaCloudcomputingplatform.Toachievethisgoal,wetrytoanticipatecomputingdemandsbypredictingaworkload,thenwemodifythesetofavailableserverstofitthispredictionandfinallywescheduleourjobsontheavailableservers.ToschedulejobswehavedevelopedthePOD(PredictOptimizeDispatch)algorithm.Weevaluateitsperformanceforreal-lifetracesinthepresenceofdifferenttypesofprediction.Theanalysisshowsthatourschedulingreducesenergyconsumptionconsiderably.
Learn-as-you-gowithMegh:EfficientLiveMigrationofVirtualMachines DebabrotaBasu(NationalUniversityofSingapore),XiayangWang(InstituteofParallelandDistributedSystems,ShanghaiJiaoTongUniversity),YangHong(ShanghaiJiaoTongUniversity),HaiboChen(ShanghaiJiaoTongUniversity),StephaneBressan(NationalUniversityofSingapore)
Weproposeareinforcementlearningalgorithm,Megh,forlivemigrationofvirtualmachinesthatsimultaneouslyreducesthecostofenergyconsumptionandenhancestheperformance.Meghlearnstheuncertaindynamicsofworkloadsasit-goes.Meghusesadimensionalityreductionschemetoprojectthecombinatoriallyexplosivestate-actionspacetoapolynomialdimensionalspace.TheseschemesenableMeghtobescalableandtoworkinreal-time.WeexperimentallyvalidatethatMeghismorecost-effectiveandtime-efficientthantheMadVMandMMTalgorithms.
Machine-LearningBasedPerformanceEstimationforDistributedParallelApplicationsinVirtualizedHeterogeneousClusters SeontaeKim(UNIST),NguyenPham(UNIST),WoongkiBaek(UNIST),Young-RiChoi(UNIST)
Inavirtualizedheterogeneouscluster,foradistributedparallelapplicationwhichrunsinmultiplevirtualmachines(VMs)concurrently,thereareahugenumberofpossiblewaystoplaceitsVMs.Thispaperinvestigatesaperformanceestimationtechniquefordistributedparallelapplicationsinvirtualizedheterogeneousclusters.WefirstanalyzetheeffectsofdifferentVMconfigurationsontheperformanceofvariousdistributedparallelapplications.Wethenpresentamachinelearningbasedperformancemodelforadistributedparallelapplication.Usingaheterogeneousclusterwithtwodifferenttypesofnodes,weshowthatourmachine-learningbasedmodelscanestimatetheruntimesofdistributedparallelapplicationswithmodesterrorrates.
IncrementalelasticityforNoSQLdatastores AntonisPapaioannou(ICS-FORTHandUniversityofCrete),KostasMagoutis(ICS-FORTHandUniversityofIoannina)
ElasticityactionsinNoSQLdatastoresmovelargeamountsofdataoverthenetworktotakeadvantageofnewresources.Hereweproposeincrementalelasticity,anewmechanismforschedulingdatatransferstoajoiningserver,leadingtosmootherelasticityactionswithareducedperformanceimpact.
AFrameworkforEfficientEnergySchedulingofSparkWorkloads StathisMaroulis(AthensUniversityofEconomicsandBusiness),NikosZacheilas(AthensUniversityofEconomicsandBusiness),VanaKalogeraki(AthensUniversityofEconomicsandBusiness)
NowadaysdistributedprocessingframeworkslikeApacheSparkhavebeensuccessfullyusedfortheexecutionofbigdataapplications.Despitetheirwideadoptionlittleworkhasbeendoneintermsofcontrollingtheapplications’energyconsumption.Datacenterscontributeover2%ofthetotalUSelectricusagethereforeminimizingtheenergyutilizationofSparkapplicationcanbeextremelyhelpful.SolvingthisenergyconsumptionproblemrequirestheschedulingofSparkapplicationsinanenergy-efficientway.However,theproblemischallengingaswealsohavetoconsiderapplicationperformancerequirements.Inthiswork,weprovidetheoverviewofanovelframeworkthatorchestratestheexecutionorderofSparkapplications,exploitingDVFStotunethecomputingnodesCPUfrequenciesinordertominimizetheenergyconsumptionandsatisfyapplication’sperformancerequirements.Ourearlyexperimentalresultsillustratetheworkingandbenefitsofourframework.
TowardsaCompleteVirtualDataCenterEmbeddingAlgorithmusingHybridStrategy MPGilesh(NationalInstituteofTechnologyCalicut),SDMadhuKumar(NationalInstituteofTechnologyCalicut),LillykuttyJacob(NationalInstituteofTechnologyCalicut),UmeshBellur(IndianInstituteofTechnologyBombay)
VirtualDataCenters(VDCs)areasetofvirtualmachine(VM)endpointsconnectedbyavirtualnetwork(VN)topology.WhileVMsarespecifiedbytheircapacityalongtheaxesofCPU,memory,andothermachinelevelresources,theVNischaracterizedbytheresourcesofvirtualswitches,andthebandwidthandlatencyoflinks.Today’sclouddatacenterssupportdynamicrequestsforVDCs,byusingsoftwaredefinedembeddingstrategies,thatallowthemtomeshmultipleVDCsontotheirphysicaldatacenter(PDC)networkandmachines.Overaperiodoftime,entriesandexitsofdifferentVDCscreatemultiplefragmentsofPDCresources,which
![Page 116: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/116.jpg)
116
areunusableunlessconsolidated,resultinginpooracceptancerateofVDCs.However,suchaconsolidationwouldnecessitatecostlyliveVMandvirtualswitchmigrations.AgoodstrategyshouldtradeoffthecostofmigrationsfortheacceptancerateofVDCrequests.Inthispaper,wepresentasolutiontothisVDCembeddingproblemthatachievesahigheracceptanceratebyminimizingfragmentation,comparedtoexistingstrategies,whileminimizingthemigrationsofVMsintheexistingVDCs.Experimentalresultsshowthatwecanachieveupto5%higheracceptancetoexistingsolutionswhilehaving6%fewerrateofmigrations.
FederatingConsistencyforPartition-ProneNetworks BenjaminBengfort(UniversityofMaryland),PeteKeleher(UniversityofMaryland)
Groupsofstronglyconsistentdevicescanefficientlyordereventsunderideal(datacenter)conditions,butbecomelesseffectiveindynamicandheterogeneousenvironments.Weaklyconsistentdevicesefficientlytoleratebothfaultsanddynamicconditionsbutareslowtoconvergeonasingleorderingofsystemevents.Wepropose“federatedconsistency”,whichcombinesthestrengthsofbothapproachesintoasingleprotocol.Federatedgroupsuseastronglyconsistentinnercoreofdevicestomaintainatotallyordered,fault-tolerantsequenceofevents.Acloudofweakly-consistentdevicesdisseminatesorderingsandenablesprogressdespitevaryingconnectivityandpartitions.Thoughtheconstituentsub-protocolstakedifferent(nearlyopposite)approachestoresolvingconflicts;weshowthatexpandingdistributedversionvectorswithafortecomponentallowsthemtointer-operateeffectively.Weuseadiscreteeventsimulationtoshowthatagroupoffederateddevicescanobtainthekeyadvantagesofbothapproaches.Suchsystemshavebeeninvestigatedbefore[1],[2],butourapproachtargetsmoreactive“weaknodes”inawide-areasetting.
Mitigatingnesting-agnostichypervisorpoliciesinderivativeclouds ChandraPrakash(IITBombay),Prashanth(IITBombay),PurushottamKulkarni(IITBombay),UmeshBellur(IITBombay)
ThefixedgranularityofvirtualmachinesofferedbyIaaSprovidershaspromptedtheevolutionofderivativecloudswhereresourcesarerepackagedintosmallercontainersandleasedouttypicallyinPaaSmode.Insuchasetup,containersareprovisionedwithinvirtualmachines.Suchanestedsetupresultsintwocontrolcentersfortheresourcesusedbythosecontainers—theguestOSandtheHypervisor.Thelatter’scontrolactionsareagnosticoftheapplicationexecutingwithinaVM.Thislackofvisibilitymayresultinhypervisorcontrolthathasanon-uniformeffectontheVM’snestedcontainerswhichisundesirable.Inthiswork,weproposepolicybasedcontroloftheeffectofthehypervisor’scontrolactionsamongstthecontainersnestedintheaffectedVM.
ANovelArchitectureforEfficientFogtoCloudDataManagementinSmartCities AmirSinaeepourfard(UPC),JordiGarcia(UPC),XavierMasip-Bruin(UPC),EvaMarin-Tordera(UPC)
Traditionalsmartcityresourcesmanagementrelyoncloudbasedsolutionstoprovideacentralizedandrichsetofopendata.Theadvantagesofcloudbasedframeworksaretheirubiquity,(almost)unlimitedresourcescapacity,costefficiency,aswellaselasticity.However,accessingdatafromthecloudimplieslargenetworktraffic,highdatalatencies,andhighersecurityrisks.Alternatively,fogcomputingemergesasapromisingtechnologytoabsorbtheseinconveniences.Theuseofdevicesattheedgeprovidesclosercomputingfacilities,reducesnetworktrafficandlatencies,andimprovessecurity.Wehavedefinedanewframeworkfordatamanagementinthecontextofsmartcitythroughaglobalfogtocloudmanagementarchitecture;inthispaperwepresentthedataacquisitionblock.Asafirstexperimentweestimatethenetworktrafficduringdatacollection,andcompareitwithatraditionalrealsystem.Wealsoshowtheeffectivenessofsomebasicdataaggregationtechniquesinthemodel,suchasredundantdataeliminationanddatacompression.
Networklet:ConceptandDeploymentShengZhang(NanjingUniversity),YuLiang(NanjingUniversityofPostsandTelecommunications),ZhuzhongQian(NanjingUniversity),MingjunXiao(UniversityofScienceandTechnologyofChina),JieWu(TempleUniversity),FanyuKong(AntFinancial),SangluLu(NanjingUniversity)
Intoday’sdatacenters,resourcerequestsfromtenantsareincreasinglytransformingintohybridrequeststhatmaysimultaneouslydemandIaaS,PaaS,andSaaSresources.Thispapertacklesthechallengeofmodelinganddeployinghybridtenantrequestsindatacenters,forwhichwecoin“networklet”torepresentasetofVMsthatcollaborativelyprovideaPaaSorSaaSservice.Throughextractingnetworkletsfromtenantrequestsandthussharingthembetweentenants,wecanachieveawin-winsituationfordatacenterprovidersandtenants.
OptimisticCausalConsistencyforGeo-ReplicatedKey-ValueStores KristinaSpirovska(EPFL),DiegoDidona(EPFL),WillyZwaenepoel(EPFL)
Inthispaperwepresentanewapproachtoimplementingcausalconsistencyingeo-replicateddatastores,whichwecallOptimisticCausalConsistency(OCC).Theoptimisminourapproachliesinthatupdatesfromaremotedatacenterareimmediatelymadevisibleinthelocaldatacenter,withoutcheckingiftheircausaldependencieshavebeenreceived.Serversperformthedependencycheckneededtoenforcecausalconsistencyonlyuponservingaclientoperation,ratherthanonthereceiptofareplicateddataitemasinexistingsystems.OCCexploresanoveltrade-offinthelandscapeofcausalconsistencyprotocols.ThepotentiallyblockingbehaviorofOCCmakesitvulnerabletonetworkpartitions.Becausenetworkpartitionsarerareinpractice,however,OCCchoosestotradeavailabilitytomaximizedatafreshnessandreducethecommunicationoverhead.WefurtherproposearecoverymechanismthatallowsanOCCsystemtofallbackonapessimisticprotocoltocontinueoperatingevenduringnetworkpartitions.POCCisanimplementationofOCCbasedonphysicalclocks.WeshowthatOCCimprovesdatafreshness,whileofferingcomparableorbetterperformancethanitspessimisticcounterpart.
AutomatedPerformanceEvaluationforMulti-TierCloudServiceSystemsSubjecttoMixedWorkloads XudongZhao(ShandongUniversity),LizhenCui(ShandongUniversity),JiweiHuang(BeijingUniversityofPostsandTelecommunications),ShijunLiu(ShandongUniversity),LeiLiu(ShandongUniversity),CaltonPu(GeorgiaTech)
Inmulti-tiercloudservicesystems,performanceevaluationreliesonnumerousexperimentsinordertocollectkeymetricssuchasresourcesusage.Theapproachmayresultinhighlytime-consuminginpractice.Inthispaper,weproposeanautomatedframeworkforperformancetracking,datamanagementandanalysistominimizehumaninterventioninmultitiercloudservicesystems.Theframeworksupportfine-grainedanalysisofthemixedworkloadsthroughtheDiscrete-timeMarkov-modulatedPoissonprocess(DMMPP).Ageneralmultitierapplicationistheoreticallyformulatedasaqueueingnetworktoevaluatetheperformance.TheeffectivenessofthemodelhasbeenvalidatedthroughextensiveexperimentsconductedintheRUBiSbenchmarksystem.
DecentralisedRuntimeMonitoringforAccessControlSystemsinCloudFederations
![Page 117: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/117.jpg)
117
MdSadekFerdous(UniversityofSouthampton),AndreaMargheri(UniversityofSouthampton),FedericaPaci(UniversityofSouthampton),MuYang,VladimiroSassone(UniversityofSouthampton)
Cloudfederationisanemergentcloud-computingparadigmwherepartnerorganisationssharedataandserviceshostedontheirowncloudplatforms.Inthiscontext,itiscrucialtoenforceaccesscontrolpoliciesthatsatisfydataprotectionandprivacyrequirementsofpartnerorganisations.However,duetothedistributednatureofcloudfederations,theaccesscontrolsystemalonedoesnotguaranteethatitsdeployedcomponentscannotbecircumventedwhileprocessingaccessrequests.Inordertopromoteaccountabilityandreliabilityofadistributedaccesscontrolsystem,wepresentadecentralisedruntimemonitoringarchitecturebasedonblockchaintechnology.
Poster4:DistributedSystemsandNetworkingCluster
DuoFS:AnAttemptatEnergy-SavingandRetainingReliabilityofStorageSystems ShuYin(HunanUniversity)
AsissuesoftheEnergyWallandtheReliabilityWallbecomeunavoidable,itisademandingandchallengingtasktoreduceenergyconsumptioninlarge-scalestoragesystemsinmoderndatacentreswhileretainingacceptablesystemsreliability.Mostenergyconservationtechniquesinevitablyhaveadverseimpactsontheparalleldisksystems.Toaddressthereliabilityissuesofenergy-efficientparallelstoragesystems,weproposeareliableenergy-efficientstoragesystemcalledDuoFS,whichaimsatimprovingbothenergyefficiencyandreliabilityofparallelstoragesystemsbyseamlesslyintegratingHDDsandSSDs.Withthehelpofthemiddlewarelayer,DuoFScandistributepopulardatatoSSD-basednodesandputHDD-basednodesintothelow-powermodeunderlightworkloadconditionswithoutmodificationoftheparallelsystems
AProposalofanEfficientTrafficMatrixEstimationunderPacketDrops KoheiWatabe(NagaokaUniversityofTechnology),ToruMano(NTTNetworkInnovationLaboratories),KimihiroMizutani(NTTNetworkInnovationLaboratories),OsamuAkashi(NTTNetworkInnovationLaboratories),KenjiNakagawa(NagaokaUniversityofTechnology),TakeruInoue(NTTNetworkInnovationLaboratories)
Trafficmatrix(TM)estimationhasbeenextensivelystudiedfordecades.Althoughconventionalestimationtechniquesassumethattrafficvolumesareunchangedbetweenoriginsanddestinations,packetsareoftendiscardedonapathduetotrafficburstiness,silentfailures,etc.ThispaperproposesanovelTMestimationmethodthatworkscorrectlyevenunderpacketdrops.ThemethodisestablishedonaBooleanfaultlocalizationtechnique;thetechniquerequiresfewercountersthoughitonlydetermineswhethereachlinkishealthy.ThispaperextendstheBooleantechniquesoastodealwithtrafficvolumeswitherrorboundsjustbyasmallnumberofcounters.Alongwithsubmodularoptimizationfortheminimumcounterplacement,weevaluateourmethodwithrealnetworkdatasets.
StragglerMitigationforDistributedBehavioralSimulation EmanBinKhunayn(UniversityofMelbourne),ShanikaKarunasekera(UniversityofMelbourne),HairuoXie(UniversityofMelbourne),KotagiriRamamohanarao(UniversityofMelbourne)
Runninglarge-scalebehavioralsimulationsrequireshighcomputationalpower,whichcanbeacquiredbydistributingcomputationworkloadtomultiplecomputingnodes(i.e.,workers)thatruninparallel.TheimplementationsofsuchsystemscommonlyfollowtheBulkSynchronousParallel(BSP)model.However,implementationsusingBSPusuallysufferfromthestragglerproblem,wherethedelayofanyworkerslowsdowntheentiresimulation.Theproblemusuallyoccursduetocommunicationdelaysorimbalancedworkloadamongworkers.Tomitigatethestragglerproblem,weproposeanovelparallelcomputationalmodel,calledPrioritySynchronousParallel(PSP)model.PSPexploitsdatadependenciesofparallelprocessestodeterminehighprioritydatatobecomputedandsynchronizedwhilecomputingtheremainingdata.PSPisimplementedandevaluatedusingtrafficsimulationsforthreelargecities.TheproposedtechniqueshowssignificantperformanceimprovementsovertheBSPmodel.
SupportingResourceControlforActorSystemsinAkka AhmedAbdelMoamen(UniversityofSaskatchewan),DezhongWang(UniversityofSaskatchewan),NadeemJamali(UniversityofSaskatchewan)
AlthoughtherearemodelsandprototypeimplementationsforcontrollingresourceuseinActorsystems,theyaredifficulttoimplementforproductionimplementationsofActorssuchasAkka.Thisisbecausethemessagingandschedulinginfrastructuresofruntimesystemsvarywidelyandareincreasinglycomplex.Inthispaper,wecomparetwodifferentwaysofapproximatingactor-levelcontrolsupportforAkka.Thefirstimplementationexpectsactormessagestoincludeestimatesofresourcesrequiredforprocessingthem.Thesecondimplementationsimplytracksactors’resourceusetodecidewhentheyshouldbeschedulednext.Wepresentexperimentalresultsontheperformancecostoftheseresourcecontrolmechanisms,aswellastheirimpactonresourceutilization.
ADistributedOperatingSystemNetworkStackandDeviceDriverforMulticores BMSaifAnsary(ECE,VirginiaTech),AntonioBarbalace(ECE,VirginiaTech),BinoyRavindran(ECE,VirginiaTech),ThomasLazor(ECE,VirginiaTech),Ho-RenChuang(ECE,VirginiaTech)
Withtheadvancesinnetworkspeedsasingleprocessorcannotcopeanymorewiththegrowingnumberofdatastreamsfromasinglenetworkcard.MulticoreprocessorscomeatarescuebuttraditionalSMPOSes,whichintegratethesoftwarenetworkstack,scaleonlytoacertainextent,limitinganapplication’sabilitytoservemoreconnectionswhileincreasingthenumberofcores.Ontheotherhand,kernelbypasssolutionsseemtoscalebetter,butlimitresourceflexibilityandcontrol.WeproposeattackingtheseproblemswithadistributedOSdesign,usingmultiplenetworkstacks(oneperkernel)andrelyingonmulti-queuehardwareandhardwareflowsteering.Thiscreatesasingle-socketabstractionamongkernelswhileminimizinginter-corecommunication.Weintroduceourdesign,consistingofadistributednetworkstack,adistributeddevicedriver,andaload-balancingalgorithm.Wecompareourprototype,NetPopcorn,withLinux,AffinityAccept,FastSocket.NetPopcornacceptsbetween5to8timesmoreconnectionsandreducesthetaillatencycomparedtothesecompetitors.WealsocompareNetPopcornwithmTCPandobservethatforhighcorecounts,mTCPacceptsonly18%moreconnectionsyetwithhighertaillatencythanNetPopcorn.
CachePotentialityofMONs:APrime PeiyanYuan(HenanNormalUniversity),HonghaiWu(HenanUniversityofScienceandTechnology),XiaoyanZhao(HenanNormalUniversity),ZhengnanDong(HenanNormalUniversity)
![Page 118: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/118.jpg)
118
NodebuffersizehasabiginfluenceonperformanceofMobileOpportunisticNetworks(MONs).Thisismainlybecauseeachnodeshouldtemporarilycachepacketstodealwiththeintermittentlyconnectedlinks.Inthispaper,westudyfundamentalboundsonnodebuffersizebelowwhichthenetworksystemcannotachievetheexpectedperformance.Giventheconditionthateachlinkhasthesameprobabilityptobeactive,andqtobeinactiveduringeachtimeslot,thereexitsacriticalvaluepcfromapercolationperspective.Ifp>pc,thenetworkisinthesupercriticalcase,thereisanachievableupperboundonthebuffersizeofnodes,independentoftheinactiveprobabilityq.Whenp<pc,thenetworkisinthesubcriticalcase,andthereexistsaclosed-formsolutionforbufferoccupation,whichisindependentofthesizeofthenetwork.
Oak:User-TargetedWebPerformance MarcelFlores(NorthwesternUniversity),AlexanderWenzel(NorthwesternUniversity),AleksandarKuzmanovic(NorthwesternUniversity)
Webperformancehaslongprovedtobeoneofthemostsoughtafteranddifficulttoachievecomponentsfortheweb.Sincetheinceptionofthemodernwebinfrastructure,thesituationhasbeengrowingincomplexity,addingremotehostsandobjects,providingeverythingfromcomputationinfrastructure,contentdistributioncapability,andtargetedadvertising.Whilemanyofthesecomponentsprovideimprovementsforsomeusers,thecomplexityoftheInternetoftenleavesotheruserssufferingfrompoorperformance.WeproposeOak,asystemwhichaddressesclientperformanceontheindividuallevel,henceaddressingchallengeswhichmaybeuniquetotheuser.Oakmeasuresauser’sperformanceforobjectsloadingonapage,anddetermineswhichcomponentsareunderperforming.Oakfurtherprovidesanautomatedmechanismbywhichsitesareabletoreplaceresourceswiththoseprovidedbyabetterperformingalternativeserviceforaparticularuser.Inthiswork,wedemonstratetheprevalenceofunder-performingservicesontheweb,findingthatover60%oftheAlexaTop500haveatleastoneunder-preformingserver.WefurtherevaluateOakonexperimentalandpopularexistingwebpages,anddemonstrateitseffectivenessinmakingdecisionsinexistingenvironmentsandwithadistributeduserbase.
Ctrl-A:ASelf-*DistributedandIn-bandSDNControlPlaneMarcoCanini(UniversitécatholiquedeLouvain),IosifSalem(ChalmersUniversityofTechnology),LironSchiff(TelAvivUniversity),EladMichaelSchiller(ChalmersUniversityofTechnology),StefanSchmid(AalborgUniversity&TUBerlin)
Adoptingdistributedcontrolplanesiscriticaltowardsensuringhighavailabilityandfault-toleranceofdependableSoftware-DefinedNetworks(SDNs).However,designingandbootstrappingadistributedSDNcontrolplaneisachallengingtask,especiallyiftobedonein-band,withoutadedicatedcontrolnetwork,andwithoutrelyingonlegacynetworkingprotocols.Oneofthemostappealingandpowerfulnotionsoffault-toleranceisself-organizationandthispaperdiscussesthepossibilityofselforganizingalgorithmsforin-bandcontrolplanes.
![Page 119: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/119.jpg)
119
Tutorial Abstracts
Tutorial1:ServerlessProgramming(FunctionasaService)PaulCastro(IBMT.J.WatsonResearchCenter),VatcheIshakian(BentleyUniversity),VinodMuthusamy(IBMT.J.WatsonResearchCenter),AleksanderSlominski(IBMT.J.WatsonResearchCenter)
ServerlessComputing(FunctionasaService)isemergingasanewandcompellingparadigmforthedeploymentofcloudapplications,largelyduetotherecentshiftofenterpriseapplicationarchitecturestocontainersandmicroservices.
FromtheperspectiveofanInfrastructure-as-a-Service(IaaS)customer,thisparadigmshiftpresentsbothanopportunityandarisk.Ontheonehand,itprovidesdeveloperswithasimplifiedprogrammingmodelforcreatingcloudapplicationsthatabstractsawaymost,ifnotall,operationalconcerns;itlowersthecostofdeployingcloudcodebychargingforexecutiontimeratherthanresourceallocation;anditisaplatformforrapidlydeployingsmallpiecesofcloudnativecodethatrespondstoevents,forinstance,tocoordinatemicroservicecompositionsthatwouldotherwiserunontheclientorondedicatedmiddleware.Ontheotherhand,deployingsuchapplicationsinaserverlessplatformischallengingandrequiresrelinquishingtotheplatformdesigndecisionsthatconcern,amongotherthings,quality-of-service(QoS)monitoring,scaling,andfault-toleranceschemes.
Fromtheperspectiveofacloudprovider,serverlesscomputingprovidesanadditionalopportunitytocontroltheentiredevelopmentstack,reduceoperationalcostsbye_cientoptimizationandmanagementofcloudresources,andenablingaserverlessecosystemthatencouragesthedeploymentofadditionalcloudservices.
Serverlessplatformspromisenewcapabilitiesthatmakewritingscalablemicroserviceseasierandcosteffective,positioningthemselvesasthenextstepintheevolutionofcloudcomputingarchitectures.MostoftheprominentcloudcomputingprovidersincludingAmazon,IBM,Microsoft,andGooglehaverecentlyreleasedserverlesscomputingcapabilities.Therearealsoseveralopen-sourceeffortsincludingtheOpenLambdaproject.
Inthistutorial,wewillpresentserverlesscomputing,surveyexistingserverlessplatformsfromindustry,academia,andopensourceprojects,identifykeycharacteristicsandusecases,anddescribetechnicalchallengesandopenproblems.Ourtutorialwillinvolveahands-onexperienceofusingtheserverlesstechnologiesavailablefromdifferentcloudproviders(e.g.IBM,Amazon,GoogleandMicrosoft).Weexpectouruserstohavebasicknowledgeofprogrammingandbasicknowledgeofcloudcomputing.
PaulCastro,Ph.D.isaResearchStaffMemberattheIBMWatsonResearchCenter.Hehasbeenactiveinresearchonmobileandpervasivecomputing,cloudinfrastructure,wirelesslocationsystems,locationdatabases,streamprocessing,andenterprisewebapplicationsandhasbeenawardedseveralpatentsintheseareas.Hehasworkedoncloudservicesforsupportingmobileapplicationsrunningonvarioussmartphoneplatforms.Workfromhisresearchintheareaofmulti-deviceapplicationsupportwasrecentlyreleasedaspartoftheIBMBluemixMobileBackendasaService.HehasearnedtwoIBMTechnicalAchievmentAwardsfortheIBMSmartCloudWebMeetingsformobileclientsandtheIntelligentNotificationSystem.Mostrecently,heworkedonIBMOpenWhiskforBluemix,withafocusonmobilesolutions.
VatcheIshakianisanAssistantProfessorintheComputerInformationSystemsdepartmentatBentleyUniversity,beforestartinghisacademiccareer,VatchewasaResearchStaffMemberatIBMResearchworkingonseveralprojectsincludingIBMOpenWhiskserverlesscomputingplatform.VatcheComputerhisPhDinComputerSciencefromBostonUniversity.Hisresearchinterestsincludedistributedbusinessprocessmanagement,Servicescomposition,andpricedbasedmodelsforcloudservices.
VinodMuthusamyisaResearchStaffMemberintheComponentSystemsGroupattheIBMT.J.WatsonResearchCenter.HecompletedhisPhDinComputerEngineeringattheUniversityofToronto.Vinod’sresearchinterestsincludepublish/subscribeeventprocessing,anddistributedbusinessprocessmanagement.Mostrecently,heworkedonIBMOpenWhiskServerlessComputingplatform.
AleksanderSlominskiisaResearchStaffMemberintheServicesandAPIEcosystemsGroupattheIBMT.J.WatsonResearchCenter.HeisinterestedindevelopmentofapplicationsforforfutureAPIEconomythattakeadvantageofupcomingcloudprogrammingapproaches,suchasserverlesscomputings,forcompositionsandorchestrationofcomponentsintobusinessworkflows.Mostrecently,heworkedonIBMIBMOpenWhiskServerlessComputingplatform.
Tutorial2:SensorCloud:ACloudofSensorNetworksSanjayMadria(MissouriUniversityofScienceandTechnology)
Traditionalmodelofcomputingwithwirelesssensorsimposesrestrictionsonhowefficientlywirelesssensorscanbeusedduetoresourceconstraints.NewermodelsforinteractingwithwirelesssensorssuchasInternetofThingsandSensorCloudaimtoovercometheserestrictions.Inthistutorial,Iwilldiscusssensorcloudarchitectures,whichenabledifferentwirelesssensornetworks,spreadinahugegeographicalareatoconnecttogetherandbeusedbymultipleusersatthesametimeondemandbasis.Iwillfurtherdiscusshowvirtualsensorsassistincreatingamultiuserenvironmentontopofresourceconstrainedphysicalwirelesssensorsandcanhelpinsupportingmultipleapplicationson-demandbasis.Iwillthenpresentsomesecurityissuesandprovideoverviewofthesolutionstotheproblemsfromtheliterature.Inparticular,Iwilldiscussenergyefficientprivacyanddataintegritypreservingdataaggregationalgorithm,riskassessmentinsensorcloudaswellasattribute-basedaccesscontrolforsensorcloudapplications.Thetopicscoveredwillbe:1.CloudofSensors–SensorCloudArchitectures2.VirtualizationinSensorCloud3.SchedulingandQoSinSensorCloud4.DatacompressionandSecureAggregationinSensorCloud5.Security,PrivacyandRiskIssuesinSensorCloud
SanjayKMadriareceivedhisPh.D.inComputerSciencefromIndianInstituteofTechnology,Delhi,Indiain1995.HeisafullprofessorintheDepartmentofComputerScienceattheMissouriUniversityofScienceandTechnology(formerly,UniversityofMissouri-Rolla,USA).Hehaspublishedover235Journalandconferencepapersintheareasofmobileandsensorcomputing,cloudandcybersecurity.HewonfiveIEEEbestpapersawardsinconferencessuchasIEEEMDM2011,IEEEMDM2012andIEEESRDS2015.Heisaco-authorofabookpublishedbySpringerinNov2003.Hehaspresentedtutorialsintheareasofsecuresensorcloud,cloudcomputing,mobilecomputing,etc.NSF,NIST,ARL,ARO,AFRL,DOE,Boeing,HangsoftandBoeinghavefundedhisresearchprojects,amongothers.HewasawardedJSPS(JapaneseSocietyforPromotionofScience)visitingscientistfellowshipin2006andASEE(AmericanSocietyofEngineeringEducation)fellowshipfrom2008to2017.In2012,hewasawardedNRCFellowshipbyNationalAcademies.Hereceivedfacultyexcellenceandresearchawardsintheyears2007,2009,2011,2013and2015fromhisuniversityforexcellenceinresearch.HeisACMDistinguishedScientist,andACMDistinguishedSpeakerandIEEESeniorMemberaswellasIEEEGoldenCoreAwardee.
![Page 120: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/120.jpg)
120
![Page 121: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/121.jpg)
121
ADSN 2017 Workshop Abstracts
UnderstandingandImprovingTemporalFairnessonanElectronicTradingVenueHaydenMelton(DeakinUniversity)
Fairness,ingeneral,isatopicthathasreceivedmuchattentioninresearchondistributedsystems.Intheirapplicationaselectronictradingvenues,however,temporalfairnessremainsatopicthatispoorlyunderstood.Thisisconcerningbecauseoperatorsofthesevenuesgenerallyhaveobligationstoensuretheirfairness.Consequently,thispaper(1)describeswhattemporalfairnessisandisnot,(2)identifiesthingsthatcanmakeitelusive,and(3)describesamechanismforimprovingitthatwasrecentlyretrofittedtoamajorFXtradingvenue:ThomsonReutersMatching.
CertificateLessCryptography-basedRuleManagementProtocolforAdvancedMissionDeliveryNetworksJonghoWon(PurdueUniversity),AnkushSingla(PurdueUniversity),ElisaBertino(PurdueUniversity)
AssuredMissionDeliveryNetwork(AMDN)isacollaborativenetworktosupportdata-intensivescientificcollaborationsinamulti-cloudenvironment.Eachscientificcollaborationgroup,calledamission,specifiesasetofrulestohandlecomputingandnetworkresources.SecurityisanintegralpartoftheAMDNdesignsincetherulesmustbesetbyauthorizedusersandthedatageneratedbyeachmissionmaybeprivacy-sensitive.Inthispaper,weproposeaCertificateLesscryptography-basedRule-managementProtocol(CL-RP)forAMDN,whichsupportsauthenticatedruleregistrationsandupdateswithnon-repudiation.WeevaluateCL-RPthroughtest-bedexperimentsandcompareitwithotherstandardprotocols.
FaultySensorDataDetectioninWirelessSensorNetworksUsingLogisticalRegressionTianyuZhang(UniversityofHyogo),QianZhao(UniversityofHyogo),YukikazuNakamoto(UniversityofHyogo)
Wirelesssensornetworks(WSNs)arecommonlyusedtomonitorchangesinanenvironmentandpreventdisasterssuchasstructuralinstability,forestfires,andtsunamis.WSNsshouldrapidlyrespondtochangesandmustprocessandanalyzesensordatainadistributedwaytominimizebatteryconsumption.Ontheotherhand,machinelearning(ML)algorithmsisapowerfultoolfordataanalyzing.However,MLalgorithmsaresocomplexthatcannotbeexecutedonresourceconstrainedsensornodes.AnotherchallengeofusingMLalgorithmsinWSNsisthatMLalgorithmsaredifficulttobedistributedoneverysensornode.BecauseMLalgorithmsarebasedonstatistics'methodsthatneedcollectingamountofdatatoapproachaccuracy.Inthispaper,weproposeamethodthatdividesalogisticalregressionMLmethodintotwosteps,thendistributesthetwostepsontosinknodeandsensornodestodetectfaultysensordata.
AnAdaptability-EnhancedRoutingMethodforMultipleGateway-basedWirelessSensorNetworksUsingSecureDispersedDataTransferRyumaTani(HiroshimaCityUniversity),KentoAoi(HiroshimaCityUniversity),EitaroKohno(HiroshimaCityUniversity),YoshiakiKakuda(HiroshimaCityUniversity)
Inconventionalwirelesssensornetworks(hereinafterreferredtoasWSNs),thesinglesinknodemodelhasbeenemployedtocollectandstorethemeasureddatatoprovidetotheexternalusersofWSNs.However,thesinglesinkmodelofWSNscanbethesinglepointoffailureforsomeusage.Tocounterthisproblem,wecanemploymultiplegateway-basedWSNs.Inaddition,WSNsaresusceptibletovariouskindsofattackssuchaseavesdropping.Tocountereavesdropping,wealreadyhaveproposedthesecretsharingscheme-basedsecuredisperseddatatransfermethod(hereinafterreferredtoasthesecuredisperseddatatransfermethod).Whilewehadconfirmedthatthesecuredisperseddatatransfermethodiseffectivetocountereavesdroppingthroughtheuseofradioareadisjointmultiplepaths,wealsofoundthatthesecuredisperseddatatransfermethodcannotbeeffectiveinsevereenvironmentssuchasinanetworkwithalowdensityofnodes.
AProgressiveDownloadMethodBasedonTimer-DrivenRequestingSchemesUsingMultipleTCPFlowsonMultiplePathsHiroakiHoriba(HiroshimaCityUniversity),TokumasaHiraoka(HiroshimaCityUniversity),JunichiFunasaka(HiroshimaCityUniversity)
Duetothewidespreaduseofbroadbandcommunicationmedia,theconventionalTCPcannotfullyutilizesuchbroadbandwidth,somanyimprovementsonTCPitselfandalotofacceleratingmethodswhichusemultipleTCPflowshavebeenproposed.Inaddition,videohostingservicesontheInternetasanewmediumhavebecomepopular,andprogressivedownloadingmethods,whichdownloadsegmentedvideodatawhilereplayingthem,areadoptedonvarioussites.TheplaybackqualityofprogressivedownloadmethodshasbeenimprovedbytheexistingmethodwhichestablishesmultipleTCPflowsoneachofmultiplepaths.However,theexistingmethodassumesthatbandwidth,delay,andpacketlossrateofeachpathareknown.Therefore,inthispaper,amethodusingthetimer-drivenrequestingschemewhichistobeeffectiveevenwhenbandwidth,delay,andpacketlossratearenotgiven.Moreover,itfeaturesduplicaterequestingschemetocopewithqualitydeteriorationinvideoplaybackduetoout-of-orderblockarrivalswhenapplyingprogressivedownloadusingmultiplepaths.Thispaperevaluatestheproposedmethodcomparingwiththeexistingmethodbysimulation.Asaresult,itisfoundthattheproposedmethodyieldshighperformanceenoughtokeepthevideoqualityhigherthantheexistingmethodeventhoughthenetworkconditionisnotclarifiedinadvance.Theproposalcanberegardedasanassurancenetworktechnologysinceitcanadapttothecurrentnetworkstatusandkeeptheplaybackratehigh.
![Page 122: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/122.jpg)
122
BGP 2017 Workshop Abstracts
WolfPath:Acceleratingiterativetraversing-basedgraphprocessingalgorithmsonGPUHuanzhouZhu(UniversityofWarwick),LigangHe(UniversityofWarwick)
Thereisthesignificantinterestnowadaysindevelopingtheframeworksofparallelizingtheprocessingforthelargegraphssuchassocialnetworks,Webgraphs,etc.Mostparallelgraphprocessingframeworksemployiterativeprocessingmodel.However,bybenchmarkingthestate-of-artGPU-basedgraphprocessingframeworks,weobservedthattheperformanceofiterativetraversing-basedgraphalgorithms(suchasBreadFirstSearch,SingleSourceShortestPathandsoon)onGPUislimitedbythefrequentdataexchangebetweenhostandGPU.Inordertotackletheproblem,wedevelopaGPU-basedgraphframeworkcalledWolfPathtoacceleratetheprocessingofiterativetraversing-basedgraphprocessingalgorithms.InWolfPath,theiterativeprocessisguidedbythegraphdiametertoeliminatethefrequentdataexchangebetweenhostandGPU.Toaccomplishthisgoal,WolfPathproposesadatastructurecalledLayeredEdgelisttorepresentthegraph,fromwhichthegraphdiameterisknownbeforethestartofgraphprocessing.InordertoenhancetheapplicabilityofourWolfPathframework,agraphpreprocessingalgorithmisalsodevelopedinthisworktoconvertanygraphintotheformatoftheLayeredEdgelist.WeconductedextensiveexperimentstoverifytheeffectivenessofWolfPath.TheexperimentalresultsshowthatWolfPathachievessignificantspeedupoverthestate-of-artGPU-basedin-memoryandout-of-memorygraphprocessingframeworks.
ANovelAuction-basedQueryPricingSchemaXingwangWang(JilinUniversity),XiaohuiWei(JilinUniversity),ShangGao(JilinUniversity),YuanyuanLiu(JilinUniversity),ZongpengLi(UniversityofCalgary)
Asacommonprocessingmethod,queryiswidelyusedinmanyareas,suchasgraphprocessing,machinelearning,statistics,etc.However,queriesareusuallypricedaccordingtovendor-specifiedfixedviews(API)ornumberoftransactions,whichignorestheheterogeneityofqueries(computingresourceconsumptionforqueryandinformationthattheanswerbrings)andviolatesthemonotoneprinciple.
Inthisworkwestudytherelationalquerypricingproblembytakingbothinformation(i.e.,data)valueandqueryresourceconsumptionsintoaccount.Wedesignefficientauctionsforquerypricing.Differentfromtheexistingquerypricingschemas,queryauctiondeterminesdatapricesthatreflectthedemand-supplyofsharedcomputingresourcesandinformationvalue(i.e.,pricediscovery).Wetargetqueryauctionthatrunsinpolynomialtimeandachievesnear-optimalsocialwelfarewithagoodapproximationratio,whileelicitstruthfulbidsfromconsumers.Towardsthesegoals,weadaptthepostedpricingframeworkingame-theoreticperspectivebycastingthequeryauctiondesignintoanIntegerLinearProgrammingproblem,anddesignaprimal-dualalgorithmtoapproximatetheNP-hardoptimizationproblem.Theoreticalanalysisandempiricalstudiesdrivenbyreal-worlddatamarketbenchmarkverifytheefficiencyofourqueryauctionschema.
Asacommonprocessingmethod,queryiswidelyusedinmanyareas,suchasgraphprocessing,machinelearning,statistics,etc.However,queriesareusuallypricedaccordingtovendor-specifiedfixedviews(API)ornumberoftransactions,whichignorestheheterogeneityofqueries(computingresourceconsumptionforqueryandinformationthattheanswerbrings)andviolatesthemonotoneprinciple.
Inthisworkwestudytherelationalquerypricingproblembytakingbothinformation(i.e.,data)valueandqueryresourceconsumptionsintoaccount.Wedesignefficientauctionsforquerypricing.Differentfromtheexistingquerypricingschemas,queryauctiondeterminesdatapricesthatreflectthedemand-supplyofsharedcomputingresourcesandinformationvalue(i.e.,pricediscovery).Wetargetqueryauctionthatrunsinpolynomialtimeandachievesnear-optimalsocialwelfarewithagoodapproximationratio,whileelicitstruthfulbidsfromconsumers.Towardsthesegoals,weadaptthepostedpricingframeworkingame-theoreticperspectivebycastingthequeryauctiondesignintoanIntegerLinearProgrammingproblem,anddesignaprimal-dualalgorithmtoapproximatetheNP-hardoptimizationproblem.Theoreticalanalysisandempiricalstudiesdrivenbyreal-worlddatamarketbenchmarkverifytheefficiencyofourqueryauctionschema.
BlockGraphChi:EnablingBlockUpdateinOut-of-coreGraphProcessingZhiyuanShao(HuazhongUniversityofScienceandTechnology),ZhenjieMei(HuazhongUniversityofScienceandTechnology),XiaofengDing(HuazhongUniversityofScienceandTechnology),HaiJin(HuazhongUniversityofScienceandTechnology)
Inthepastseveralyears,lotsofout-of-coregraphprocessingsystemsarebuilttoprocessbiggraphdatasetsincomputersystemswithlimitedmainmemory.Duetotheiterativenatureofgraphalgorithms,mostofthesesystemsemploysynchronousexecutionmodeltoorganizethecomputation,i.e.,dividethecomputingintomultiplerounds,eachofwhichcorrespondstooneiterationofthegraphalgorithm.Inordertofullyutilizethediskbandwidth,thesesystemssequentiallyscanthewholegraphdatasetateachiteration.However,asthegraphdatasetunderprocessingmaybehuge,moreiterationsgenerallymeanslargerI/Ooverheads.Althoughasynchronousimplementationofthesynchronousexecutionmodelallowsmessagepassingwithinaniteration,theeffectivenessisstilllimited.Sinceinsuchmodel,atmostonemessageisallowedtobepassedfromonevertextoanother.
Inthispaper,weinvestigatetheideaofblockupdatinginthesynchronousexecutionmodelframeworkintheout-of-coregraphprocessingsystems.Withthisnewmodel,thesystemconductsgraphalgorithmontheloadedsubgraph(i.e.,block)toitslocalconvergence,andthenswitchestoothersubgraphstocontinuethisprocess,untiltheglobalconvergenceisreached.WeimplementthisnewmodelinGraphChi(theresultsystemiscalledBlockGraphChi),andproposeagraphpartitionmethod,namedasDMLP,tocooperatewiththisnewmodel.Bythisstudy,wefoundthatcomparedwiththeoriginalexecutionmodelofGraphChi:1)thenewmodelcangenerallyreducetheamountofiterations(andthustheI/Ooverheads)forgraphalgorithms,whiletheextentofreductiondependsonthemethodofgraphpartitioningandthepropertiesofthealgorithms;2)thenewmodelcandramaticallyreducetheoverallexecutiontimeofgraphtraversalalgorithms(byupto31.4x),andbetterpartitioningmethodleadstobetterperformance;3)thenewmodelhasmuchsmallereffectivenessonimprovingtheoverallperformanceoffix-pointalgorithms,suchasPageRank,duetotheincreasedcomputationaloverhead.
IncrementalParallelComputingusingTransactionalModelinLarge-scaleDynamicGraphStructuresAnandTripathi(UniversityofMinnesota,Minneapolis),RahulR.Sharma(UniversityofMinnesota),ManuKhandelwal(UniversityofMinnesota),TanmayMehta(UniversityofMinnesota),VarunPandey(UniversityofMinnesota)
Manygraphanalyticsproblemsbenefitfromtheuseofparallelcomputingtechniquestoreducetheexecutiontime,whichcanstillbequitehighforlargegraphproblems.Thegoalofourworkistoeliminatetheneedofre-executingananalyticsprogramwhenthegraphstructureismodifiedwithasmallsetofupdatesaftertheinitialexecutionoftheprogram.Towardsthisgoal,wepresentheretheresultsofourinvestigationofincrementalcomputationtechniquesindynamicgraph
![Page 123: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/123.jpg)
123
structuresusingatransactionalmodelofparallelprogramming.Inthismodel,computationtasksinananalyticsapplicationareexecutedinparallelasserializabletransactions.Thispaperdescribeshowincrementalcomputationtechniquesaresupportedbythismodelfordynamicgraphstructures.Weusetheproblemsoffindingconnected-componentsinagraphandthegraphcoloringproblemtoillustrateourapproachforincrementalcomputations.Usingexperimentalevaluations,weshowthebenefitsofthisapproach.
AgainstSigned-GraphDeanonymizationAttacks:PrivacyProtectionforSocialNetworksJianliangGao(CentralSouthUniversity),YuLiu(CentralSouthUniversity),PingZhong(CentralSouthUniversity),JianxinWang(CentralSouthUniversity)
Socialnetworksareusuallypresentedasgraphs.Butthetopologicalcharacteristicsofgraphscouldbeusedbyattackerstodeanonymizetargetentitiesinsocialnetworks.Existingworksmostlyhaveanassumptionthatattackerknowsonlythetargetentities'neighborhoodgraph.Thisassumptionmightresultinprivacyleakagebecauseoftheignoranceoflinkpropertybetweenentities.Inrealapplications,attackersmightre-identifyentitiesinsocialnetworksbasedonnotonlythelinksbetweenentities,butalsothepropertyoflinks.Inthispaper,wetakethepropertyoflinksintoconsiderationforthefirsttimewhenachieving$k$-anonymityforsocialnetworks,whichmeanstheattackerscannotre-identifyatargetwithconfidencehigherthan$1/k$.Thelinksarecatalogedaspositiveandnegative,whichiscalledsignedgraph.Inthisbackground,weproposea$k$-anonymizationschemetoprotecttheprivacyofkeyentitiesinsocialnetworks.Theproposedschememinimizestheamountofmodificationonoriginalgraphs,whichpreservestheutilityoftheoriginaldata.Extensiveexperimentsonrealdatasetsandsyntheticgraphillustratetheeffectivenessoftheproposedscheme.Theutilityofanonymizednetworksareremainedbydemonstratingwiththeresultsofvertexdegree,betweenness,closenessandtheirKolmogorov-Smirnov(K-S)test.
![Page 124: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/124.jpg)
124
CCN-CPS 2017 Workshop Abstracts
PoliciesGuidingCohesiveInteractionsamongInternetofThingswithCommunicationCloudandSocialNetworksHenryHexmoor(SouthernIllinoisUniversity)
CohesiveinteractionamongInternetofthingnodeswillbenefitfromformationofadhoccommunicationnetworkcloudsforrapidexchangeofinformationthatispertinentfortheirsuccessfulinteraction.Longenduringinteractionsamongsuchnodeswillbenefitfromadhocsociallylinkednetworksforcollaborationonsharedobjectives.Wepresentguidelinesforformingandusingtheseconstructsandpoliciesthatconstrainthemtorequirementsofspecificapplications.
EnhancedSecurityofBuildingAutomationSystemsThroughMicrokernel-BasedControllerPlatformsXiaolongWang(UniversityofSouthFlorida),RichardHabeeb(UniversityofSouthFlorida),XinmingOu(UniversityofSouthFlorida),SiddharthAmaravadi(KansasStateUniversity),JohnHatcliff(KansasStateUniversity),MasaakiMizuno(KansasStateUniversity),MitchellLNeilsen(KansasStateUniversity),RajRajagopalan(Honeywell),SrivatsanVaradarajan(HoneywellAerospaceAdvancedTechnologyLabs)
ABuildingAutomationSystem(BAS)isacomplexdistributedCyber-PhysicalSystemthatcontrolsbuildingfunctionalitiessuchasheating,ventilation,andaircondition-ing(HVAC),lighting,access,emergencycontrol,andsoon.ThereisagrowingopportunityandmotivationforBAStobeintegratedintoenterpriseITnetworkstogetherwithvariousnew""smart""technologiestoimproveoccupantcomfortandreduceenergyconsumption.Thesenewtechnologiescoexistwithlegacyapplications,creatingamixed-criticalityenvironment.Inthisenvironment,assystemsareintegratedintoITnetworks,newattackvectorsareintroduced.Thus,networkednon-criticalapplicationsrunningontheOSplatformmaybecompromised,leavingthecontrolsystemsvulnerable.Theindustryneedsareliablecomputingfoundationthatcanprotectandisolatetheseendangeredcriticalsystemsfromuntrustedapplications.
Thisworkpresentsanovelkernel-basedapproachtosecurecriticalapplications.Ourmethodusesasecurity-enhanced,microkernelarchitecturetoensurethesecurityandsafetypropertiesofBASinapotentiallyhostilecyberenvironment.WecomparethreesystemdesignandimplementationsforasimpleBASscenario:1)usingthemicrokernelMINIX3enhancedwithmandatoryaccesscontrolforinter-processcommunication(IPC),2)usingseL4,aformallyverified,capability-basedmicrokernel,and3)usingLinux,amonolithickernelOS.Weshowthroughexperimentthatwhenthenon-criticalapplicationsarecompromisedinbothMINIX3andseL4,thecriticalprocessesthatimpactthephysicalworldarenotaffected.WhereasinLinux,thecompromisedapplicationscaneasilydisruptthephysicalprocesses,jeopardizingthesafetypropertiesinthephysicalworld.ThisshowsthatmicrokernelsareasuperiorplatformforBASorothersimilarcontrolenvironmentsfromasecuritypointofview,anddemonstratesthroughexamplehowtoleveragethearchitecturetobuildarobustandresilientsystemforBAS.
HighlevelDesignofaHomeAutonomousSystemBasedonCyberPhysicalSystemModelingBasmanAlhafidh(FloridaInstituteofTechnology),WilliamH.Allen(FloridaInstituteofTechnology)
Theprocessusedtobuildanautonomoussmarthomesystemusingcyber-physicalsystems(CPS)principleshasreceivedmuchattentionbyresearchersanddevelopers.However,therearemanychallengesduringthedesignandimplementationofsuchasystem,suchasPortability,Timing,Prediction,andIntegrity.Thispaperpresentsanovelmodelingmethodologyforasmarthomesysteminthescopeofcyber-physicalinterfacethatattemptstoovercometheseissues.Wediscussahigh-leveldesignapproachthatsimulatesthefirstthreelevelsofa5CarchitectureinCPSlayersinasmarthomeenvironment.Adetaileddescriptionofthemodeldesign,architecture,andasoftwareimplementationviaNetLogosimulationprogramwillbepresented.Ourdesignprovidesanexamplefordevelopersonhowtoimplementanecosysteminahomeenvironmentaspartofasmartcities'infrastructurebasedonCPSdesignprinciples.
ACyberPhysicalBuses-and-DronesMobileEdgeInfrastructureforLargeScaleDisasterEmergencyCommunicationsMamtaNarang(AucklandUniversityofTechnology),WilliamLiu(AucklandUniversityofTechnology),JairoAGutierrez(AucklandUniversityofTechnology),LucaChiaraviglio(UniversityofRomeTorVergata)
Immediatelyafteradisaster,thenormaltelecommunicationinfrastructure,includingwiredandwirelessnetworks,isoftenseriouslycompromisedandcannotguaranteeregularcoverageandreliablecommunicationsservices.Thesetemporarily-missingcommunicationscapabilitiesarecrucialtorescuersandaffectedcitizensastherespondersneedtoeffectivelycoordinateandcommunicatetominimizethelossoflivesandproperty.Acyber-physicalsystem(CPS)iscomposedofintegratedcommunication,computationandphysicalobjects,andcyber-physicalvehiclesystems(CPVSs)areanemergingfieldduetotherapidadvancementsonreal-timecomputing,mobilecommunicationsandautonomouscontrolinintelligenttransportsystems.Inthispaper,weproposeacyber-physicalbuses-and-dronesmobiLeedgeinfrastructure(AidLife)fordisasteremergencycommunications,whichaimsatarapidlydeployableresilientsystemcapableofsupportingflexiblecommunicationstoservelarge-scaledisastersituationsbyutilizingtheexistingpublictransportsystem.Inparticularweenvisionaproposalwherepublicbusescanberecruitedtotemporarilyhostportablebasestation(BS)andcomputationunitsaswellaspowerresourcessoastoformabuses-basedmobileedgeinfrastructure,andalsoaccommodatedronestoextendtheircoveragetohard-to-reachareas.OurpreliminaryresultsshowthattheAidLifesystemcanguaranteeagoodcoveragetousers,evenwhenalargenumberofnormalBSsthataredamaged.
APerformanceComparisonofContainersandVirtualMachinesinWorkloadMigrationContextKumarGaurav(VMwareSoftwareIndiaPvtLtd),PavanKarkun(VMwareSoftwareIndiaPvtLTD),Y.C.Tay(NationalUniversityofSingapore)
Thispapergivesamathematicalframeworkfordecisionmakingaroundplacingandmigratingworkloadsinadata-centerwhereapplicationsarepackagedasOScontainersrunningonvirtualmachines.ThedecisionpointonVMmigrationvscontainerkill/restart,VMforkvscontainerspawnarestudiedhere.Weproposeamathematicalmodelforthemigrationofworkloadsaforementionedcasesandalsoforsharedmemorydecayincaseofforkingavirtualmachine.Experimentalresultsareanalyzedtodeterminethevalidityofthemodel.
TowardsService-OrientedMiddlewareforCyberPhysicalSystemsNaderMohamed(MiddlewareTechnologiesLab.),SanjaLazarova-Molnar(UniversityofSouthernDenmark)
Cyber-PhysicalSystems(CPS)providemanysmartfeaturesforenhancingphysicalsystemsandenvironments.Theyaredesignedwithasetofdistributedhardware,software,andnetworkcomponentsthatareembeddedinphysicalsystemsandenvironmentsorattachedtohumans.ManyCPSatdifferentscalesarebeingdevelopedforavarietyofapplicationsthatprovidevaluableinteractionsbetweenthecyberworldandthephysicalsystemsandenvironments.However,thesedevelopmentsfacemanychallengesduetothecomplexityoftheseapplications.Anappropriatemiddlewareisneededtoprovideinfrastructuralsupportandassist
![Page 125: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/125.jpg)
125
thedevelopmentandoperationsofdiverseCPSapplications.Thispaperstudiesutilizingtheservice-orientedmiddlewareapproachforCPSanddiscussestheadvantagesandrequirementsforsuchutilization.Inaddition,itproposesaservice-orientedmiddlewareforCPS,calledCPSWare.ThismiddlewareviewsallCPScomponentsasasetofservicesandprovidesaninfrastructuretodevelopandoperateCPSapplications.ThisapproachprovidessystemicsolutionsforsolvingmanycomputingandnetworkingissuesinCPS.ItalsoenablestheintegrationofCPSwithothersystemssuchasCloudComputingandFogComputing.Inaddition,asCPScanbedevelopedforvariousapplicationsatdifferentscales,thispaperprovidesaclassificationforCPSapplicationsanddiscusseshowCPSWarecaneffectivelydealwiththesecategories.
NetworkingandCommunicationinCyberPhysicalSystemsImadJawhar(UAEUniversity),JameelaAl-Jaroodi(RobertMorrisUniversity)
Cyber-physicalsystems(CPSs)areemergingasanewtechnology,whichisusedtoprovideseamlessinteractionbetweenthephysicalandcyberworlds.Thisnovelparadigmisanaturalevolutionandextensionofwirelesssensornetworks(WSNs)andcontrolmodelstoallowforeffectivemonitoringandcontrolofphysicalsystemsfromthecomputingenvironment.Inordertosupportthisinterfaceandallowsuchsmoothinteractions,efficientnetworkingandcommunicationbetweenthephysicalandcyberworldstakeaveryimportantandcriticalrole.Inthispaper,weidentifythevariousapplicationsandcategoriesofCPSsystems,andcharacterizetheassociateddatatrafficthatisgenerated.Wealsodiscussthedifferentprotocolsandrequirementsthatareneededatthevariousnetworkinglayersfortheseapplications.Subsequently,weidentifyimportantparameterssuchasbandwidth,delay,reliability,security,andmobility,whichareessentialinordertoallowforeffectiveandrobustoperationofthevariousCPSsystems.
OptimalDeploymentofChargingStationsforElectricVehicles:AFormalApproachAmarjitDatta(TennesseeTechnologicalUniversity),BrianLedbetter(TennesseeTechnologicalUniversity),MohammadAshiqurRahman(TennesseeTechnologicalUniversity)
Electricvehicles(EVs)areafascinatinginnovationofthemodernautomobileindustry.Duetotheirattractivefeaturesandagrowingworldwideenvironmentalawareness,thenumberofEVpurchasesisgrowingatanincreasingratedaybyday.AsthepriceofEVsisexpectedtodropinthenearfuture,alargenumberofnewEVswillhittheroadconsequently.However,ourcurrentinfrastructureisnotcapableofsupportingthisgrowingnumberofEVs.Weneedmorechargingstations,placedoptimallyacrossanarea,eachequippedwithmultiplechargingoutletstochargetheincomingEVsinareasonableamountoftime.Inthispaper,wepresentaformalframeworktooptimallydeploychargingstationsforEVsinagivenarea.TheframeworkdesignsthisverificationasanoptimizationproblemwherethegoalistooptimallyplacethechargingstationswithasufficientnumberofchargingoutletstoserveallEVsinagivenareawhilesatisfyingthelimitedbudgetandothersystemconstraints.Weevaluatetheproposedframeworkforitsanalysiscapabilityaswellasitsscalabilitybyexecutingexperimentsondifferentsynthetictestcases.
FormalVerificationofControlStrategiesforaCyberPhysicalSystemAmjadGawanmeh(KhalifaUniversityofScienceandTechnology),AliAlwadi(AucklandUniversityofTechnology),SaziaParvin(UniversityofNewSouthWales)
CyberPhysicalSystems(CPS)useemergingcomputing,communication,andcontrolmethodstomonitorandcontrolgeographicallydispersedcriticalsystemcomponentstoallowahighlevelofconfidenceabouttheiroperation.Simulationmethodsarefrequentlyusedintestingsuchcriticalsystemcomponents,however,itmightnotbeadequatetoshowtheabsenceoferrorsgiventhecomplexityofthesystemcomponentsundertest.Failureindetectingerrorsinsafetycriticalsystemscanleadtoacatastrophicsituation.Inthispaperweproposeanapproach,basedonsimulationandformalanalysis,forthereliabilityanalysisofCPS.WeillustratethisapproachonanindustrialcasestudythatdemonstratesseveralchallengingfeaturesinthedesignandimplementationofCPS.Experimentalresultsobtainedshowthattheproposedapproachisefficientlyusedinordertotestandverifythefourtanksprocesssystem,wheresimulationresultsshowthevalidityofapproximationandabstractionofthesystem,andformalanalysisisusedtovalidatethatseveraldesignrequirementsweresatisfiedinthecontrolstrategiesproposed.
LightweightDetectionandIsolationofBlackHoleAttacksinConnectedVehiclesSamiAlbouq(OaklandUniversity),ErikFredericks(OaklandUniversity)
ConnectedVehicles(CVs)canbeexposedtoblackholeattacksthatdeceivelegitimatenodesbyfalsifyinganattractiveroutetoadestinationnode.Thisoccurswhenanattackersendsapackettothesourcenodeconfirmingtheexistenceofafreshroute.Inthispaper,weproposeaBlackHoleDetectionProtocol(BlackDP)thatworksonahighwaydividedintoclustersandmonitoredbyRoadSideUnits(RSUs)todetectbothsingleandcooperativeblackholeattacks.EveryRSUistaskedwithperformingbothdetectionandisolationofblackholeattacksfortheirrespectivehighwaysectionafterauthenticationviolationsandsuspiciousrouteestablishmentactivitiesthathavebeenreportedbyalegitimatenode.ThedesigngoalofBlackDPistodecouplethedetectionprocessfrommobilenodesandconstructatrustedsemi-centricdetectionprocessthatcancollectneededinformationforlightweightdetectionandreliableisolationofmaliciousnodes.WevalidateBlackDPinasimulatedhighwayenvironmenttodemonstrateitseffectiveness.
AnewthreatassessmentmethodforintegratinganIoTinfrastructureinaninformationsystemBrunoDorsemaine(OrangeLabs),Jean-PhilippeGaulier(OrangeLabs),Jean-PhilippeWary(OrangeLabs),NizarKheir(Thales),PascalUrien(TelecomParisTech)
Inthispaper,weproposeanewapproachtomanagethethreatsbroughtbyanIoTinfrastructuretoaninformationsystem(IS).WefirstgiveastateofartforinformationsecuritypropertiesinIoTandISbasedonstandardssuchasISO16982andISO27005andapreviouslypublishedtaxonomy.Thenwedetailaninnovativemethod,basedontheevaluationofthreatsbroughtbyanIoTinfrastructureontoanIS.ItisrepresentedasaqualitativematrixbetweenIoTinfrastructurethreatsandtheSecuritypropertiesoftheIS.Themethodisthenappliedtotheusecaseofconnectedlightbulbs.Thankstothisapproach,itispossibletologicallyorganizethreatmanagementwhileintegratinganIoTinfrastructureintoanIS.
ASecurityFrameworkforSDN-enabledSmartPowerGridsUttamGhosh(TennesseeStateUniversity),PushpitaChatterjee(SRMRESEARCHINSTITUTE),SachinShetty(OldDominionUniversity)
Emergingsoftwaredefinednetworking(SDN)paradigmprovidesflexibilityincontrolling,managing,anddynamicallyreconfiguringsmartgridnetworks.ItcanbeseenintheliteraturethatconsiderablylessattentionhasbeengiventoprovidesecurityinSDN-enabledsmartgridnetworks.Mostoftheeffortsfocusonprotecting
![Page 126: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/126.jpg)
126
smartgridnetworksagainstvariousformsofoutsiderattacksonlybyprovidingconsistentaccesscontrol,applyingefficientandeffectivesecuritypolicies,andmanagingandcontrollingthenetworkthroughtheuseofacentralizedSDNcontroller.Furthermore,centralizedSDNcontrollersareplaguedbyreliabilityandsecurityissues.ThispaperpresentsaframeworkwithmultipleSDNcontrollersandsecuritycontrollersthatprovidesasecureandrobustsmartgridarchitecture.TheproposedframeworkdeploysalocalIDSinasubstationtocollectthemeasurementdataperiodicallyandtomonitorthecontrol-commandsthatareexecutedonSCADAslaves.AglobalIDSincontrolcentercollectsthemeasurementdatafromthesubstationsandestimatesthestateofthesmartgridsystembyutilizingthetheoryofdifferentialevolution.TheglobalIDSfurtherverifiestheconsequencesofcontrol-commandsissuedbySDNcontrollerandSCADAmaster.Analarmisgeneratedupondetectionofanattackerorunsteadystateofthesmartgridsystem.Theframeworkalsodeployslight-weightidentitybasedcryptographytoprotectthesmartgridnetworkfromoutsideattacks.PerformancecomparisonandinitialsimulationresulthavebeenpresentedtoshowthattheproposedframeworkiseffectiveascomparedtoexistingsecurityframeworksforSDN-enabledsmartgrids.
Real-timeMonitoringSteamGeneratorsusingaHybridImagingSystemMahmoudMeribout(PetroleumInstitute),ImranSaied(PetroleumInstitute),EsraAlHosani(AdcoGroup)
Thispaperpresentsahybriddeviceforreal-timemeasurementandimagingofsolidandliquidcontaminantsthatmayoccurinsteamgenerators.ThedeviceusesadedicatedNearInfra-Reddevicetodeterminethetypeofcontaminants(i.e.waterdropletsandironoxideparticles)andaTHzimagingsystemwhichmeasurestheamountofcontaminantsaswellasitsflowrate.TheNIRdevicecanalsodeterminetheconcentrationofcontaminantsatsub-mgaccuracywhenitsvalueisrelativelylowusingspectrometrytechniquecombinedwithprincipalcomponentanalysis(PCA).Threeprincipalcomponents(PC1,PC2,andPC3)wereenoughforthispurpose.ThePCAclassificationwasperformedusingtheleastsquaresupportvectormachine(LS-SVM)method.Incaseofrelativelyhighconcentration,theTHzimagingsystemwhichusesblock-basedmotionestimationalgorithmcandeterminethevelocityofindividualcontaminantparticlestocomputetheglobalmotionvector,theintensityanddirectionofwhichrepresentstheoverallflowrateandflowregimeofthecontaminants.TheusageofimageprocessingtechniquestogetherwithNIRspectrometryconstitutesanewpromisingstepinflowmetering.ThisisdemonstratedbytheextensiveexperimentswhichhavebeenconductedfordifferentscenariowheretheNIRsubsystemsystemcoulddeterminetheconcentrationofwaterdropletsandsolidcontaminantswithamaximumuncertaintyof+/-1.45%and+/-1.16%respectively.WiththeNIRsubsystem,pixel-levelaccuracyofmotionvectorwasachieved,whiletheconcentrationofsolidcontaminantsshowedconsistedproportionalitywiththeaveragepixelintensity.
SecuringbigDataEfficientlythroughMicroaggregationTechniqueandHuffmanCompressionShakilaMahjabinTonni(BangladeshArmyInternationalUniversityofScienceandTechnology),MohammadZahidurRahman(JahangirnagarUniversity),SaziaParvin(UniversityofNewSouthWales),AmjadGawanmeh(KhalifaUniversityofScienceandTechnology)
Cyber-PhysicalSystems(CPS)requiresbigdatacommunicationsaswellasintegrationfromseveraldistributedsources.Thisdatacanusuallybeinterconnectedwithphysicalapplications,suchaspowergridsorSCADAsystems.Inaddition,itcanbepubliclyaccessibleforusingbythirdpartyusersordatascientists.Therefore,itbecomesimperativetoensurethatthisbigdataiswellsecured.Microaggregationisanwidelyusedtechniquetoprotectadatasetthroughanonymityinordertopreventexposureofaperson'sidentity.Thisdatadisclosuremayalsoresultfromanunpredicteddatalinkagewithanotherdataset.As,mostofthesesurveydatasetsstorerecordsusingnumericalvalues,manyofthemicroaggregationtechniquesaredevelopedandtestedonnumericaldata.Thesealgorithmsarenotsuitableforthosedatawherebothnumericalandcategoricaldataarestored.Inthispaperwe'reproposingamicroaggregationtechniqueinordertoprovidedataanonymityregardlessofitstype.TherecordsareclusteredintoseveralgroupsusinganevolutionaryattributegroupingalgorithmandeachgrouprecordsarethenmicroaggregatedapplyingHuffmandatacompressionalgorithm.
ModelBasedEnergyConsumptionAnalysisofWirelessCyberPhysicalSystemsJingLiu(PekingUniversity),PingWang(PekingUniversity),JinlongLin(PekingUniversity),Chao-HsienChu(PennsylvanniaStateUniversity)
Wirelessmeshnetworksbegintobeusedasaninfrastructureofcyber-physicalsystems.Acriticalissueindevelopingwirelesscyberphysicalsystems(WCPSs)isthelimitedamountofenergyavailableinthenodes.Energyconsumptionanalysiscanhelpdesignertoconductapower-awaredesignprocess.Inthispaper,weproposeamodelbasedenergyconsumptionanalysisframeworkatarchitecturelevelforWCPSs.Weextracteventchainsfromthearchitecturemodel,withtheenergyconsumptionmodelforprocessingeachtypeofevent,wecanestimatetheenergyconsumptionforeachcontrolloopandeachnode,aswellastheoverallenergyconsumption.AlltheseenergyconsumptionindexescanhelpustodesignaperformanceandenergyconsumptionbalancedWCPS.
![Page 127: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/127.jpg)
127
HotPOST 2017 Workshop Abstracts
Router-basedBrokeringforSurrogateDiscoveryinEdgeComputingJulienGedeon(TechnischeUniversitätDarmstadt),ChristianMeurisch(TechnischeUniversitätDarmstadt),DishaBhat(TechnischeUniversitätDarmstadt),MichaelStein(TechnischeUniversitätDarmstadt),LinWang(TechnischeUniversitätDarmstadt),MaxMühlhäuser(TechnischeUniversitätDarmstadt)
In-networkprocessingpushescomputationalcapabilitiesclosertotheedgeofthenetwork,enablingnewkindsoflocation-aware,real-timeapplications,whilepreservingbandwidthinthecorenetwork.Thisisdonebyoffloadingcomputationstomorepowerfulorenergy-efficientsurrogatesthatareopportunisticallyavailableatthenetworkedge.Inmobileandheterogeneoususagecontexts,thequestionariseshowaclientcandiscoverthemostappropriatesurrogateinthenetworkforoffloadingatask.Inthispaper,weproposeabrokeringmechanismthatmatchesaclientwiththebestavailablesurrogate,basedonspecifiedrequirementsandcapabilities.Thebrokerisimplementedonstandardhomerouters,andthus,leveragestheubiquityofsuchdevicesinurbanenvironments.Tomotivatethefeasibilityofthisapproach,weconductacoverageanalysisbasedoncollectedaccesspointlocationsinamajorcity.Furthermore,thebrokeringfunctionalityintroducesonlyaminimalresourceoverheadontheroutersandcansignificantlyreducethelatencycomparedtodistant,cloud-basedsolutions.
ModelingtheSpreadofInfluenceforIndependentCascadeDiffusionProcessinSocialNetworksZeshengChen(IndianaUniversity-PurdueUniversityFortWayne),KurtisTaylor(IndianaUniversity-PurdueUniversityFortWayne)
Modelingthespreadofinfluenceinonlinesocialnetworksisimportantforpredictingtheinfluenceofindividualsandbetterunderstandingmanyscenariosinsocialnetworks,suchastheinfluencemaximizationproblem.Thepreviousworkonmodelingthespreadofinfluencemakestheassumptionthatthestatusesofnodesinanetworkareindependentofeachother,whichisapparentlynotcorrectforsocialnetworks.Thegoalofthisworkistoderiveanaccuratemathematicalmodeltocharacterizethespreadofinfluencefortheindependentcascadediffusionprocessinonlinesocialnetworks.Specifically,weapplythesusceptible-infected-recoveredepidemicmodelfromepidemiologytocharacterizetheindependentcascadediffusionprocessandderiveageneralmathematicalframework.Toapproximatethecomplexspatialdependenceamongnodesinanetwork,weproposeaMarkovmodeltopredictthespreadofinfluence.Throughtheextensivesimulationstudyoverseveralgeneratedtopologiesandarealcoauthorshipnetwork,weshowthatourdesignedMarkovmodelhasmuchbetterperformancethantheexistingindependentmodelinpredictingtheinfluenceofindividualsinonlinesocialnetworks.
ThankYouForBeingAFriend:AnAttackerViewonOnline-Social-Network-basedSybilDefensesDavidKoll(UniversityofGoettingen),MartinSchwarzmaier(UniversityofGoettingen),JunLi(UniversityofOregon),Xiang-YangLi(UniversityofScienceandTechnologyofChina),XiaomingFu(UniversityofGoettingen)
OnlineSocialNetworks(OSNs)havebecomearewardingtargetforattackers.OneparticularlypopularattackistheSybilattack,inwhichtheadversarycreatesmanyfakeaccountscalledSybilsinorderto,forinstance,distributespamormanipulatevotingresults.AfirstgenerationofdefensesystemstriedtodetecttheseSybilsbyanalyzingchangesinthestructureoftheOSNgraph---unfortunatelywithlimitedsuccess.Basedontheseeffortsasecondgenerationofsolutionsenrichesthegraph-structuralapproacheswithhigher-leveluserfeaturesinordertodetectSybilnodesmoreefficiently.Inthisworkweprovideanin-depthanalysisofthesedefenses.Wedescribetheircommondesignandworkingprinciples,analyzetheirvulnerabilities,anddesignsimpleyeteffectiveattackstrategiesthatanadversarycouldlaunchtocircumventthesesystems.InourevaluationwerevealthatanmiscreantcanexploitthecredulityofOSNusersandfollowatargetedattackstrategytosuccessfullyavoiddetectionbyallexistingapproaches.
EfficientDynamicServiceFunctionChainCombinationofNetworkFunctionVirtualizationWenkeYan(BeijingUniversityofPostsandTelecommunications),KonglinZhu(BeijingUniversityofPostsandTelecommunications),LinZhang(BeijingUniversityofPostsandTelecommunications),SixiSu(BeijingUniversityofPostsandTelecommunications)
NetworkFunctionVirtualization(NFV)andSoftwareDefinedNetwork(SDN)arerecentlyintroducedtoprovidethevirtualizationtechnologyfortacklingthedeploymentofnetworkservicefunctionsincorporatenetworks,broadbandaccessnetworks,andmorerecentlyindatacenters.Howtoenhancetheflexibility,efficiencyandeffectiveofservicefunctiondeploymentisfullofchallenge.AlthoughServiceFunctionChain(SFC)iscarriedouttosupporttheflexibilityofnetworkservices,itstillneedsonestepforwardtofulfilltheefficientandeffectivecombinationofnetworkservices.Inthispaper,weproposeanorthogonalcrossoverdifferentialevolution(OXDE)tooptimizeSFCcombinationwithrespecttoprocessingdelay,energyconsumption,andpacketlossrate.TheevaluationresultsshowthattheproposedOXDEalgorithmoutperformstheotheralgorithmsanditcanachievetheefficiencyandeffectivenessofSFCcombination.
WhenAugmentedRealitymeetsBigDataCarlosBermejo(TheHongKongUniversityofScienceandTechnology),ZhanpengHuang(TheHongKongUniversityofScienceandTechnology),TristanBraud(TheHongKongUniversityofScienceandTechnology),PanHui(TheHongKongUniversityofScienceandTechnology)
Weliveinanerawhereweareoverloadedwithdata,andthiscanbethekeyforgainingrichinsightsaboutourworld.Augmentedreality(AR)enablesusthepossibilitytovisualiseandanalysethegrowingtorrentofdatainainteractive,usablecanvas.Wecandisplaycomplexdatastructuresinsimplerandmoreunderstandablewaysthatwasnotpossiblebefore.BigDataisanewparadigmresultsfromthemyriaddatasourcessuchastransactions,Internet,socialnetworks,healthcaredevicesandsensornetworks.ARandbigdatahavealogicalmaturitythatinevitablywillconverge.ThetreadofharnessingARandbigdatatobreednewinterestingapplicationsisstartingtohaveatangiblepresence.Inthispaper,weexplorethepotentialtocapturevaluefromthemarriagebetweenARandbigdatatechnologies,followingwithseveralchallengesthatmustbeaddressedtofullyrealizethispotential.
SamplingBasedEfficientAlgorithmtoEstimatetheSpectralRadiusofLargeGraphsSamarAbbas(LahoreUniversityofManagementSciences),JuvariaTariq(LahoreUniversityofManagementSciences),ArifZaman(LahoreUniversityofManagementSciences),ImdadullahKhan(LahoreUniversityofManagementSciences)
Evaluatinganextremelyusefulgraphproperty,thespectralradius(largestabsoluteeigenvalueofthegraphadjacencymatrix),forlargegraphsrequiresexcessivecomputingresources.Thisproblembecomesespeciallychallenging,forinstancewithdistributedorremotestorage,whenaccessingthewholegraphitselfisexpensiveintermsofmemoryorbandwidth.Oneapproachtotacklethischallengeistoestimatethespectralradiusofthegraphwhilereadingonlyasmallportion
![Page 128: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/128.jpg)
128
ofthegraph.Inthispaperwepresentasamplingapproachtoestimatethespectralradiusoflargegraphs.Wedefineascoreforverticesthati)ismoreofacombinatorialnatureandiseasiertocomputeandii)hassolidtheoreticaljustificationshence,itcloselyapproximateavertex'scontributiontothelargesteigenvalueofthegraph.Usingthisscore,wemodelthesamplingproblemasabudgetedoptimizationproblemanddesignagreedyalgorithmtoselectasubgraphwhosespectralradiusapproachesthatofthewholegraph.Weprovideanalyticalboundoncomputationalcomplexityofouralgorithm.Wedemonstrateeffectivenessofouralgorithmonvarioussyntheticandreal-worldgraphsandshowthatouralgorithmalsoempiricallyoutperformsknowntechniques.Furthermore,wecomparethequalityofourresultstoestimatesobtainedfromwellknownupperandlowerboundsknowninthespectralgraphtheoryliterature.
ExtemporaneousMicro-MobileServiceExecutionWithoutCodeSharingZhengSong(VirginiaTech),MinhLe(UtahStateUniversity),Young-WooKwon(UtahStateUniversity),EliTilevich(VirginiaTech)
Inmobileedgecomputing,amobileorIoTdevicerequestsanearbydevicetoexecutesomefunctionalityandreturnbacktheresults.However,theexecutablecodemusteitherbepre-installedonthenearbydeviceorbetransferredfromtherequesterdevice,reducingtheutilityorsafetyofdevice-to-devicecomputing,respectively.Toaddressthisproblem,wepresentamicro-servicemiddlewarethatexecutesservicesonnearbymobiledevices,withatrustedmiddlemandistributingexecutablecode.Oursolutioncomprises(1)atrustedstoreofvettedmobileservices,self-containedexecutablemodules,downloadedtodevicesandinvokedatruntime;and(2)amiddlewaresystemthatmatchesservicerequirementstoavailabledevicestoorchestratethedevice-to-devicecommunication.Ourexperimentsshowthatoursolution(1)enablesexecutingmobileservicesonnearbydevices,withoutrequiringadevicetoreceiveexecutablecodefromanuntrustedparty;(2)supportsmobileedgecomputinginpracticalsettings,increasingperformanceanddecreasingenergyconsumption;(3)reducesthemobiledevelopmentworkloadbyreusingservices.
PreventingColludingIdentityCloneAttacksinOnlineSocialNetworksGeorgesA.Kamhoua(FloridaInternationalUniversity),NikiPissinou(FloridaInternationalUniversity),S.S.Iyengar(FloridaInternationalUniversity),JonathanBeltran(FloridaInternationalUniversity),CharlesKamhoua(AirForceResearchLaboratory),BrandonLHernandez(UTRGV),LaurentNjilla(AirForceResearchLaboratory)
Nowadays,OnlineSocialNetworks(OSNs)becomeoneofthemostcommonwayamongstpeopletofacilitatecommunication,thishasmadeitatargetforattackerstostealinformationfrominfluentialusers.ThishasbroughtnewformsofcustomizedattacksforOSNs.AttackerstakeadvantageoftheusertrustworthinesswhenusingOSN.Thisexploitationleadstoattackswithacombinationofbothclassicalandmodernthreats.Specifically,colludingattackershavebeentakenadvantageofmanyOSNsbycreatingfakeprofilesoffriendsofthetargetinthesameOSNorothers.Colludersimpersonatetheirvictimsandaskfriendrequeststothetargetintheaimtoinfiltrateherprivatecircletostealinformation.ThistypeofattacksaredifficulttodetectinOSNsbecausemultiplemalicioususersmayhaveasimilarpurposetogaininformationfromtheirtargeteduser.Inthispaper,toovercomethistypeofattack,wefirstaddresstheproblemofmatchinguserprofilesacrossmultipleOSNs,second,byusingbothtextualandfeaturesextractedfromuserprofileandbasedonsupervisedlearningtechniques,webuildaclassifier.Simulationandexperimentalresultsareprovidedtovalidatetheaccuracyofourfindings.
![Page 129: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/129.jpg)
129
IoTCA 2017 Workshop Abstracts
TowardsPrivacy-AwareSmartBuildings:Capturing,Communicating,andEnforcingPrivacyPoliciesandPreferencesPrimalPappachan(UniversityofCaliforniaIrvine),MartinDegelingy(CarnegieMellonUniversity),RobertoYus(UniversityofCaliforniaIrvine),AnupamDasy(CarnegieMellonUniversity),SrutiBhagavatulay(CarnegieMellonUniversity),WilliamMelichery(CarnegieMellonUniversity),PardisEmamiNaeiniy(CarnegieMellonUniversity),ShikunZhangy(CarnegieMellonUniversity),LujoBauery(CarnegieMellonUniversity),AlfredKobsa(UniversityofCaliforniaIrvine),SharadMehrotra(UniversityofCaliforniaIrvine),NormanSadeh(CarnegieMellonUniversity),NaliniVenkatasubramanian(UniversityofCaliforniaIrvine)
TheInternetofThings(IoT)ischangingthewayweinteractwithoursurroundingenvironmentindomainsasdiverseashealth,transportation,officebuildingsorourhomes.Insmartbuildingenvironments,informationcapturedaboutabuilding’sinfrastructureanditsinhabitantswillhelpdevelopservicesthatcanhelpusbecomemoreproductive,increaseourcomfort,enhanceoursocialinteractions,increasesafety,saveenergyandmore.Butbyrelyingonthecollectionandsharingofinformationaboutabuilding’sinhabitantsandtheiractivities,theseservicesalsoopenthedoortoprivacyrisks.Inthispaper,weintroduceaframeworkwhereIoTAssistantscaptureandmanagetheprivacypreferencesoftheirusersandcommunicatethemtoprivacy-awaresmartbuildings,whichenforcethemwhencollectinguserdataorsharingitwithbuildingservices.Weoutlineelementsofaninfrastructurenecessarytosupportsuchinteractionsandalsodiscussimportantprivacypolicyattributesthatneedtobecaptured.Thisincludeslookingatattributesnecessarytodescribe–(1)thedatacollectionandsharingpracticesassociatedwithdeployedsensorsandservicesinsmartbuildingsaswellas(2)theprivacypreferencesweneedtocapturetohelpusersmanagetheirprivacyinsuchenvironments.
DeployingData-DrivenSecuritySolutionsonResource-ConstrainedWearableIoTSystemHangCai(WorcesterPolytechnicInstitute),TianlongYun(WorcesterPolytechnicInstitute),JosiahHester(DartmouthCollege),KrishnaK.Venkatasubramanian(ClemsonUniversity)
WearableInternet-of-Things(WIoT)environmentshavedemonstratedgreatpotentialinabroadrangeofapplicationsinhealthcareandwell-being.SecurityisessentialforWIoTenvironments.LackofsecurityinWIoTsnotonlyharmsuserprivacy,butmayalsoharmtheuser’ssafety.ThoughdevicesintheWIoTcanbeattackedinmanyways,inthispaperwefocusonadversarieswhomountwhatwecallsensorhijackingattacks,whichpreventtheconstituentmedicaldevicesfromaccuratelycollectingandreportingtheuser’shealthstate(e.g.,reportingoldorwrongphysiologicalmeasurements).Inthispaperweoutlinesomeofourexperiencesinimplementingadatadrivensecuritysolutionfordetectingsensor-hijackingattackonasecurewearableinternet-of-things(WIoT)basestationcalledtheAmulet.Giventhelimitedcapabilities(computation,memory,batterypower)oftheAmuletplatform,implementingsuchasecuritysolutionisquitechallengingandpresentsseveraltradeoffswithrespecttoresourcesrequirements.WeconcludethepaperwithalistofinsightsintowhatcapabilitiesconstrainedWIoTplatformsshouldprovidedeveloperssoastomaketheinclusionofdata-drivensecurityprimitivesonsuchsystemseasy.
AMotifbasedIoTFrameworkforDataEfficiencyAkashSahoo(TexasA&MUniversity),RabiMahapatra(TexasA&MUniversity)
InternetofThings(IoT)hasallowedembeddeddevicestoconnecttothevastInternetnetworkworldwide.WithbillionsofIoTdeviceswaitingtobeconnected,itisnecessarytobuildefficientinfrastructuretohandlelargeamountofdataforefficientstorageandnetworktraffic.TheamountofdatacreatedattheIoTedgesisregardedasonethebiggestchallengesofIoT.Thispaperproposesamotif-basedencodingschemeforIoTframeworkthathelpstoreducedatageneratedbysensorsatedgenodes.Thissimpleencodingfeatureresidesinboththeserverandtheenddeviceslikeinserver-clientmodel.Ourexperimentsdemonstratedtheschemesbenefitsbyusingslowandfastbaudratesensorssuchastemperatureandaccelerometerrespectivelyasthecasestudies.Theresultsobtainedshowtheproposedmotifbasedframeworkreducesthedataredundancyuptotwoordersofmagnitudewhileretainingmorethan80%accuracytowardsmotifrecognition.
CoTWare:ACloudofThingsMiddlewareJameelaAl-Jaroodi(RobertMorrisUniversity),NaderMohamed(MiddlewareTechnologiesLab.),ImadJawhar(MidcompResearchCenter)
Therearemanyapplicationsthatrequireintegratingalargenumberofphysicalobjectsanddevicesinalarge-scaleInternetofThings(IoT)networks.Someexamplesoftheseapplicationsaresmartgrids,smartwaternetworks,andintelligenttransportationsystems.Theseapplicationsneedreal-timecontrols,powerfulandscalabledatastorageandprocessingcapabilities,andadvanceddataanalyticsmechanisms.OneofthepromisingtechnologiestosupportsuchapplicationsistheCloudofThings(CoT).CoTcanprovideaplatformforlinkinganIoTwithCloudComputing(CC).AnothertechnologythatcanbeutilizedforenhancingIoTapplicationsisFogComputing,whichextendsthetraditionalCloudComputingparadigmtotheedgeofthenetworktoenablebettersupportforoperatingenhancedservices.However,properintegrationandefficientutilizationofCoTandFogComputingforlarge-scaleIoTapplicationsisnotaneasytask.Thispaperproposesaservice-orientedmiddleware,calledCoTWare,tofacilitateeffectiveintegrationandutilizationofCoTandFogComputingforlarge-scaleIoTapplications.
SecuringtheInternetofThings:AMeta-StudyofChallenges,Approaches,andOpenProblemsMahmudHossain(UniversityofAlabamaatBirmingham),RagibHasan(UniversityofAlabamaatBirmingham),AnthonySkjellum(AuburnUniversity)
TheInternetofThings(IoT)isbecomingakeyinfrastructureforthedevelopmentofsmartecosystems.However,theincreaseddeploymentofIoTdeviceswithpoorsecurityhasalreadyrenderedthemincreasinglyvulnerabletocyberattacks.Insomecases,theycanbeusedasatoolforcommittingseriouscrimes.AlthoughsomeresearchershavealreadyexploredsuchissuesintheIoTdomainandprovidedsolutionsforthem,thereremainstheneedforathoroughanalysisofthechallenges,solutions,andopenproblemsinthisdomain.Inthispaper,weconsiderthisresearchgapandprovideasystematicanalysisofsecurityissuesofIoT-basedsystems.Then,wediscusscertainexistingresearchprojectstoresolvethesecurityissues.Finally,wehighlightasetofopenproblemsandprovideadetaileddescriptionforeach.WepositthatoursystematicapproachforunderstandingthenatureandchallengesinIoTsecuritywillmotivateresearcherstoaddressingandsolvingtheseproblems.IndexTerms—InternetofThings;SecurityIssue;AttackSur-face;AttackTaxonomy;IoTForensics.
InternetofThingsFrameworkforSmartLearningAnalyticsAliYavari(SwinburneUniversityofTechnology),RezaSoltanpoor(RMITUniversity)
![Page 130: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/130.jpg)
130
LearningAnalytics(LA)hasbecomeaprominentparadigminthecontextofeducationlatelywhichadoptstherecentadvancementsoftechnologysuchascloudcomputing,bigdataprocessing,andInternetofThings.LAalsorequiresanintensiveamountofprocessingresourcetogeneraterelevantanalyticalresults.However,thetraditionalapproacheshavebeeninefficienttotackleLAchallengessuchasreal-time,highper-formance,andscalableprocessingofheterogeneousdatasetsandstreamingdata.AnInternetofThings(IoT)scalable,distributedandhighperformanceframeworkhasthepotentialtoaddressmentionedLAchallengesbyefficientcontextualizationofdata.Inthispaper,SmartLearningAnalyticsconceptualmodelisproposedtoimprovetheeffectivenessofLAbyutilizinganIoT-basedplatformintermsofperformance,scalability,andefficiency.
![Page 131: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/131.jpg)
131
JCC 2017 Workshop Abstracts
HeterogeneousMalwareSpreadProcessinStarNetworkLiboJiao(TsinghuaUniversity),HaoYin(TsinghuaUniversity),DongchaoGuo(TsinghuaUniversity),YongqiangLyu(TsinghuaUniversity)
TheheterogeneousSISmodelforvirusspreadinanyfinitesizegraphcharacterizestheinfluenceoffactorsofSISmodelandcouldbeanalyzedbytheextendedN-Intertwinedmodelintroducedin[1].Wespecificallyfocusontheheterogeneousvirusspreadinthestarnetworkinthispaper.Theepidemicthresholdandtheaveragemeta-stablestatefractionofinfectednodesarederivedforvirusspreadinthestarnetwork.OurresultsillustratetheeffectofthefactorsofSISmodelonthesteadystateinfection.
CostReductioninHybridCloudsforEnterpriseComputingBiyuZhou(InstituteofComputingTechnology,ChineseAcademyofSciences),FaZhang(InstituteofComputingTechnology,ChineseAcademyofSciences),JieWu(TempleUniversity),ZhiyongLiu(InstituteofComputingTechnology,ChineseAcademyofSciences)
Hybridcloud-baseddeploymentisatrendincloudcomputingwhichenablesenterprisetobenefitfromcloudinfrastructureswhilehonoringprivacyrestrictionsonsomeservices.Enterpriseapplicationmigrationisaneffectivewaytoimprovetheefficiencyofusingthecloudinfrastructures.However,itisachallengingproblemtodecidewhichpartsoftheapplicationstomigrateandwheretomigrate.Inthispaper,wefocusontheproblemofplanningthemigrationofenterpriseapplicationsinhybridcloudinfrastructures.Unlikepreviousstudies,weconsiderageneralhybridcloudarchitecturethatinvolvesmultiplepubliccloudsratherthanonlyone.Ouraimistomaximizetheenterprisecostreductionundertheconstraintofuserexperienceintermsofresponsetime.Wefirstformulatethepplicationmigrationproblemasanoptimizationproblem.AwareofitsNP-hardness,wedesignanefficientmigrationframeworktoapproximatetheoptimumforalargeproblemsize.First,weleveragetheapplicationcharacteristictoreducethescaleoftheproblembydividingitintomultiplesmallersubproblems.Then,anefficientalgorithmbasedondynamicprogrammingisproposedtosolvethesmallscalesubproblems.Finally,weconstructafeasiblesolutiontotheoriginalproblem.Simulationresultsdemonstratethatourframeworkcanbringsignificantbenefitstoenterprises.
DC-RSF:ADynamicandCustomizedReputationSystemFrameworkforJointCloudComputingFanghuaYe(SunYat-senUniversity),ZibinZheng(SunYat-senUniversity),ChuanChen(SunYat-senUniversity),YurenZhou(SunYat-senUniversity)
Jointcloudcomputing(JointCloud),asabrand-newparadigmofcloudcomputing,aimsatbuildingacloudecosystem,inwhichendusersareagnostictocloudservicevendorsasapplicationsandservicesarebuiltuponvirtualclouds.IncaseoflowqualitycloudresourcesprovideddeliberatelyandinordertofacilitatethepersistentandsounddevelopmentofJointCloudecosystem,weproposeadynamicandcustomizedreputationsystemframework(DC-RSF)toevaluatethecredibilityofcloudservicevendors.AtthecoreofDC-RSFisthecustomizedanddynamiccredibilitymodel(CDCM),whichcalculatescreditvalueforeachcloudservicevendorbasedonservicerequirementsofendusersandcredentialattributesofcloudservicevendors.WefurtherincorporateaBlockchain-basedmoduleintoDC-RSFtopreventthecreditvaluefrombeingartificiallytampered.
WebServiceApplianceBasedonUnikernelKaiYu(NationalLabforParallelandDistributedProcessing),ChengfeiZhang(NationalLabforParallelandDistributedProcessing),YunxiangZhao(NationalLabforParallelandDistributedProcessing)
Mini-OSisatinyOS(operatingsystem)kerneldistributedwithXenProjectHypervisor.ItismainlyusedasanOSforstubdomainaimedatDom0disaggregationandalsoasteppingstoneforUnikerneldevelopment.WeimplementedasimplehttpserveronMini-OS,andbuiltMini-OSintoawebserviceappliance.WeevaluateditsperformancecomparedwiththesameimplementedserveronUbuntuPV(para-virtualization)DomU,andachievedabout39%performanceimprovement.TheresultsshowsthatMini-OScanbeawebserviceapplianceandhasagoodperformance.
AnalysisandEvaluationoftheGASModelforDistributedGraphComputationWangJinyan(NationalLabforParallelandDistributedProcessing),ZhangChengfei(NationalLabforParallelandDistributedProcessing)
Comparedwithdistributedgraphcomputation,traditionallysinglenodecomputationisunfittedinprocessinglargescalegraphdata.TheGAS(Gather,ApplyandScatter)Modelisauniversalvertex-cutgraphcomputationprogrammingmodelbasedonedge-centricprogramstosupportgraphalgorithms,whichprocessdistributedgraphcomputationaftergraphpartition.Inthispaper,weintroducethatthreeminor-stepsofGAS.WethenanalyzemorecompleteprocessofGASconsideringintra-nodecomputationandinter-nodecommunicationofdistributedgraphcomputation.Basedonouranalysis,weevaluatetheperformanceindifferentnodesofgraphanalysisalgorithmapplyingGASmodel.Theevaluationshowsthatthebottleneckiscomputationperformanceorcommunicationbandwidthdependingonnumberofnodes,whichisaninspirationofoptimizingtheGASmodel.
TrafficSignsDetectionBasedonFasterR-CNNZhongrongZuo(NationalLabforParallelandDistributedProcessing),KaiYu(NationalLabforParallelandDistributedProcessing),QiaoZhou(NationalLabforParallelandDistributedProcessing),XuWang(NationalLabforParallelandDistributedProcessing),TingLi(NationalLabforParallelandDistributedProcessing)
Inthispaper,weuseaadvancedmethodcalledFasterR-CNNtodetecttrafficsigns.Thisnewmethodrepresentsthehighestlevelinobjectrecognition,whichdon'tneedtoextractimagefeaturemanuallyanymoreandcansegmentimagetogetcandidateregionproposalsautomatically.Ourexperimentisbasedonatrafficsigndetectioncompetitionin2016byCCFandUISEEcompany.ThemAPvalueoftheresultis0.3449thatmeansFasterR-CNNcanindeedbeappliedinthisfield.Eventhoughtheexperimentdidnotachievethebestresults,weexploreanewmethodintheareaofthetrafficsignsdetection.Webelievethatwecangetabetterachievementinthefuture.
JCLedger:ABlockchainBasedDistributedLedgerforJointCloudComputing
![Page 132: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/132.jpg)
132
XiangFu(NationalUniversityofDefenseTechnology),HuaiminWang(NationalUniversityofDefenseTechnology),PeichangShi(NationalUniversityofDefenseTechnology),YingweiFu(NationalUniversityofDefenseTechnology),YijieWang(NationalUniversityofDefenseTechnology)
WiththedevelopmentofEconomicGlobalization,traditionalsingle-cloudproviderscannotmeettheneedsoftheexplosive,global,diversecloudservices.JointCloudaimsatempoweringthecooperationamongmultipleCloudServiceProviders(CSP)toprovidecross-cloudservices.OurworkinthispaperismainlyfocusedontheaccountingtechnologyforJointCloudcomputingandweproposetheJCLedger-ablockchainbaseddistributedledger.AnewparticipantCCP(CryptocurrencyProvider)isintroducedintotheJointCloudcollaborationenvironmenttoprovidethecryptocurrencytransferred.WehaveadetaileddescriptionofJCLedgermodel.WefurtheranalyzethefourmostimportantmechanismsforJCLedgerandprovidebasicperspectivesforin-depthanalysis.Finally,wediscusstheinnovationsofJCLedgerandourfutureworkinthisfield.
CorporationArchitectureforMultipleCloudServiceProvidersinJointCloudComputingPeichangShi(NationalUniversityofDefenseTechnology),HuaiminWang(NationalUniversityofDefenseTechnology),XikunYue(NationalUniversityofDefenseTechnology),ShilanYang(NationalUniversityofDefenseTechnology),ShangzhiYang(NationalUniversityofDefenseTechnology),YuxingPeng(NationalUniversityofDefenseTechnology)
Nowadays,cloudcomputingishardtoeffectivelysustaintheimplementationofthecommercialmodelofInternetServiceglobalization.Thereisagrowingtrendtobuildanenvironmentofcloudservice,withthecapacitytoserveanytimeandanywhere,bymutualcooperationbetweencloudserviceprovidersaroundtheworld.However,thistendencywillraiseakeyissuewhichishowtoprovideabenignenvironment,thatallowsself-collaborationandfaircompetition,fordifferentcloudserviceproviderswithdiversestakeholder.GuidedbytheconceptandstructureofService-OrientedArchitecture(SOA)service,thispaperproposesastructurenamedJointCloudCorporationEnvironment(JCCE),whichoffersamutualbenefitandwin-winJointCloudenvironmentforglobalcloudserviceproviders.JCCEcontainsthreecoreservices,whichareDistributedCloudTransaction,DistributedCloudCommunityandDistributedCloudSupervision.Also,facingwithdifferentcloudserviceparticipants,JCCEoffersthreemainservicemodesfortheirconsumption,supplyandcoordination.Thisstudyplaysasignificantroleinsupportingthesharingandself-collaborationofmultiplecloudentities,andpromotingthedevelopmentofcloudservicemarkethealthyandorderly.
SharingPrivacyDatainSemi-TrustworthyStoragethroughHierarchicalAccessControlYuzhaoWu(TsinghuaUniversity),YongqiangLyu(TsinghuaUniversity),QianFang(TsinghuaUniversity),GengZheng(TsinghuaUniversity),HaoYin(TsinghuaUniversity),YuanchunShi(TsinghuaUniversity)
Dataoutsourcingincloudisemergingasasuccessfulparadigmthatbenefitsorganizationsandenterpriseswithhigh-performance,low-cost,scalabledatastorageandsharingservices.However,thisparadigmalsobringsforthnewchallengesfordataconfidentialitybecausetheoutsourcedarenotunderthephysiccontrolofthedataowners.Theexistingschemestoachievethesecurityandusabilitygoalusuallyapplyencryptiontothedatabeforeoutsourcingthemtothestorageserviceproviders(SSP),anddisclosethedecryptionkeysonlytoauthorizeduser.Theycannotensurethesecurityofdatawhileoperatingdataincloudwherethethird-partyservicersareusuallysemi-trustworthy,andneedlotsoftimetodealwiththedata.WeconstructaprivacydatamanagementsystemappendinghierarchicalaccesscontrolcalledHAC-DMS,whichcannotonlyassuresecuritybutalsosaveplentyoftimewhenupdatingdataincloud.
AReliabilityBenchmarkforBigDataSystemsonJointCloudYingyingZheng(InstituteofSoftware,ChineseAcademyofSciences),LijieXu(InstituteofSoftware,ChineseAcademyofSciences),WeiWang(InstituteofSoftware,ChineseAcademyofSciences),WeiZhou(KSYUN),YingDing(ChangchunUniversityofScienceandTechnology)
JointCloudprovidesaflexibleandelasticcomputingresourceplatform.BigdatasystemssuchasMapReduceandSparkarewidelydeployedonthisplatformforbigdataprocessing.Theseframeworkshavehighscalability,buttheapplicationsrunningatopthemoftengenerateruntimeerrors,suchasoutofmemoryerrors,IOexceptions,andtasktimeouts.Forusers,theywanttoknowwhetherthedevelopedapplicationshavepotentialapplicationfaults.Forsystemdesignersandmanagers,theywanttoknowwhetherthedeployed/updatedframeworkshavepotentialsystemfaults.Currentperformancebenchmarkingcanchoosesuitablecloudsplatformforcustomers.However,theydonotconsiderreliabilityofapplicationsdeployedonthecloud.Inaddition,currentbenchmarksforbigdatasystemarealsoonlydesignedforperformancetesting.Tofillthisgap,weproposeareliabilitybenchmark,whichcontainsrepresentativeapplications,anabnormaldatagenerator,andaconfigurationcombinationgenerator.Differentfromperformancebenchmarks,thisbenchmark(1)generatesabnormaltestdataaccordingtotheapplicationcharacteristics,and(2)reducestheconfigurationcombinationspacebasedonconfigurationfeatures.Currently,weimplementedthisbenchmarkonSparkframework.Inourpreliminarytest,wefoundthreetypesoferrors(i.e.,outofmemoryerror,timeoutandwrongresults)infiveSQL,MachineLearning,andGraphapplications.
UCPR:UserClassificationandInfluenceAnalysisinSocialNetworkCongZha(TsinghuaUniversity),YongqiangLv(TsinghuaUniversity)
Therearevigorousdevelopmentsofsocialnetworkwhichaffectoutlifegreatly.Userinfluenceisanimportantreasontopro-motetheinteractioninsocialnetwork.Whenweanalyzeuserinfluence,singlevaluecan’tindicatetheuserinfluenceindifferentdomains.ThispaperputsforwardthedesignofUserClassi-ficationPageRank(UCPR)tosolvethisproblem.Firstly,weclassifyusersaccordingtothecontentwhichtheyforwarded.Then,weusespacemappingtosetupseveralsubnet.Finally,weanalyzeuserinfluenceineveryspecificsubnetbyDomainMappedNetwork(DMN)whichisbasedonPageRankalgorithmandweimprovethisalgorithmtoanalyzetheuserinfluenceindifferentdomains.Throughtheworkofthispaper,weusedavectortopresentuserinfluenceratherthanasinglenumberandwetestandverifiedthelong-taileddistributionsofsocialnet-workinexperiments.
AdaptiveRoutingAlgorithmforJointCloudVideoDeliveryZexunJiang(TsinghuaUniversity),HaoYin(TsinghuaUniversity)
AstheInternetkeepsgrowing,onlinevideohasbecomeagreatpartofthecurrentInternetdatatraffic,whichwilltakeover80%ofInternettrafficaccordingtoCiscosreport.Also,newandmoreheavyweightapplicationskeepdevelopingtofulfillpeople’sgrowingrequirements,like4kresolutionandvisualrealityvideos.However,onesingleserviceprovider,likeaContentDeliveryNetwork(CDN),cannotmeettheperformancerequirementscompletely.ToemploythepotentialofJointCloud,thispaperdesignsandimplementsanewrequestroutingalgorithmthatcanmakevideodeliveryutilizemultiplecloudsandservers.Onthepremiseofguaranteeingthequalityofvideoplaying,thisalgorithmminimizesthecostofserviceresourcesbasedondifferentinfrastructuresservicequality,cost,andcover
![Page 133: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/133.jpg)
133
areas.Basedonthisalgorithm,weimplementapracticalvideodeliverysystemusinglight-weight,flash-basedterminals.Andthissystemprovideslivevideoandvideo-on-demanddeliveryserviceforChinaFutureNetworksIndustrySummit2014onJune4th.Theactualuserdatawasgatheredandanalyzedtoverifytheeffectivenessofthisalgorithm.90.2%ofthetotalVODrequestswerecompletedsmoothlywithoutpause,andthevideotrafficwasoptimizedbythealgorithm.
TowardsEfficientResourceManagementinVirtualCloudsBoAn(PekingUniversity),JunmingMa(PekingUniversity),DonggangCao(PekingUniversity),GangHuang(PekingUniversity)
Theuseofmultiplecloudsbringsmanyadvantages:costoptimization,Quality-of-Service(QoS)improvements,highavailability,avoidanceofvendorlock-in,disasterrecoveryandsoon.However,currentlythecloudvendorislargelyproprietaryanddifferentcloudvendorshavetheirownheterogeneousinfrastructure,makingitdifficultforuserstoutilizeresourcefrommultiplecloudvendors.Asaresult,usershavetomanagedistributedapplicationsspanningmultiplecloudsandtakeintoconsiderationtheservicesmigrationforreasonslikebestcostefficiency.Inthispaper,weintroducesthenotionofVirtualCloudandfocusontheissuesrelatedtomulti-cloudresourcemanagementinVirtualCloud.VirtualCloudisacustomizedcloudbyaggregatingresourcesandservicesofdifferentcloudsandaimstoprovideenduserswithaspecificcloudworkingenvironment.Itwilleaseusers’burdenofresourceanddistributedapplicationmanagementaswellastheworkloadmigrationacrosscloud.
MonitoringandBillingofALightweightCloudSystemBasedonLinuxContainerYujianZhu(PekingUniversity),JunmingMa(PekingUniversity),BoAn(PekingUniversity),DonggangCao(PekingUniversity)
Nowadays,moreandmoreenterprisesandresearchinstituteschoosetobuildmini-datacentersanddeployprivatecloudenvironmentstomeetgrowingbusinessandresearchneeds.Tomakeuserscanrundifferentapplicationframeworksonthesamedatacenter,Caoetal.proposedanewservicemodelnamedClaaS(ClusterasaService)anddevelopedalightweightprototypesystemnamedDockletwhichisbasedonLXC(LinuxContainer).Dockletfacesaproblemofresourceswasteandabuseduetoourfreepolicy.ThispaperintroducesthemonitoringandbillingmodulesofDockletinordertosolvethisproblem.Monitoringmoduleprovidesusersandadministratorswithaclear,real-timeanddetailedmonitoringinterfacetounderstandthestatuesofrunningapplicationsandtheusageofphysicalresources.Billingmoduleusesthesedatatoreminduserstoreleaseunnecessaryresources.Anexperimentandobservationsshowthatourproposedmonitoringmethodiseffectiveandlightweightandourproposedbillingmodelincreasestheutilizationofphysicalresourcesofamini-datacenter.
Buildingemulationframeworkfornon-volatilememoryGuoliangZhu(NationalUniversityofDefenseTechnology),KaiLu(NationalUniversityofDefenseTechnology),XiaopingWang(NationalUniversityofDefenseTechnology)
Currently,researchersusesimulatorstoexperimenttheirinnovationonemergingnon-volatilememory.Unfortunately,simulationmethodisbothtime-consumingandarehardtodebug.Inthispaper,wepresentanon-volatilememoryemulatorwhichenablessystem-levelresearchonemergingmemory.Ouremulatorusesperformancemonitoringunitsonoff-the-shelfprocessorstoimplementanaccureteperformancemodel.
Seflow:EfficientFlowSchedulingforData-ParallelJobsQiaoZhou(NationalLabforParallelandDistributedProcessing),ZiyangLi(NationalLabforParallelandDistributedProcessing),PingZhong(CentralSouthUniversity),TianTian(NationalLabforParallelandDistributedProcessing),YuxingPeng(NationalLabforParallelandDistributedProcessing)
Data-paralleljobstransfermassiveamountsofdatabetweenaseriesofsuccessivestages.Thecoflowabstractionisproposedtorepresentagroupofparallelflowsbetweentwostagesandefficientlyimprovesstagelevelperformance.However,state-of-the-artcoflowschedulingtechniquesareagnostictothejobs’intercoflowsemanticsandthusaresuboptimalinreducingtheaveragejobcompletiontimes(JCT).Toaddressthisproblem,inthispaperwepresentthe“semanticflow”(seflow)abstractiontoexpressthejob-levelintercoflowsemantics.Aseflowcomprisesnotonlyallthecoflowsofajobbutalsotherelationshipbetweenthecoflows.Wedesignanefficientseflowschedulerwhichutilizestherichseflowsemanticsofjobstoachievebetterperformancethanseflow-agnosticschedulingfordataparalleljobs.
OnlineEncodingforErasure-CodedDistributedStorageSystemsFangliangXu(NationalUniversityofDefenseTechnology),YijieWang(NationalUniversityofDefenseTechnology),XingkongMa(NationalUniversityofDefenseTechnology)
Manylarge-scaledistributedstoragesystemsdeployerasurecodingtoprotectdatafromfrequentserverfailuresforcostreason.Inmostofthesesystems,newlyinserteddataisfirstreplicatedacrossdifferentstoragenodesandthenmigratedtoerasurecoded.Althoughthisofflineencodingmannercanimproveperformanceofdataaccessbeforeerasurecodingforsomesystems,ithelpslittleandwastesmanynetworkresourcesanddiskresourcesformanyothersystems.Inthisstudy,weproposeanonlineencodingmethod,whichencodesdataassoonasitisinsertedintothesystem.Byeliminatingthemigrationprocess,ouronlineencodingcansignificantlyreducenetworktransferanddataread;bycachingtheintermediateparityblocksintomemory,ouronlineencodingalsosignificantlyreducedatawrite.Analysisshowthatouronlineencodingcanreducedatatransferbymorethan25%,reducedatawriteby57%atleastandeliminatealldataread,comparedtotraditionalofflineencoding.
![Page 134: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/134.jpg)
134
PED 2017 Workshop Abstracts
WED-SQL:ARelationalFrameworkforDesignandImplementationofProcess-AwareInformationSystemsBrunoPadilha(UniversityofSaoPaulo),AndréLuisSchwerz(FederalUniversityofTechnology),RafaelLiberatoRoberto(FederalUniversityofTechnology)
Despitethesignificantevolutionofthedesignandimplementationofbusinessprocessmodels,atransactionalapproachthatevolvesanincrementalandadaptivestrategyremainsanimportantchallengetobeovercome.TraditionalframeworkssuchasBPEL,ProcessAlgebra,andPetriNetrequireanadditionalsoftwarelayerorsomethirdpartytoolkitstobeabletoenforceadata-statebasedtransactioncontrolanddealwithsemanticexceptions.However,thecomplexityofimplementationbasedonthesetraditionalframeworks,especiallytotreatexceptions,istoohigh.Inthispaper,wepresenttheWED-SQL,adistributedframeworkthatprovidesareliableandefficientwaytodesignandimplementbusinessprocesses.OurmaincontributionistheintegrationofWED-flowconceptsintothePostgresSQLRDBMS.ThisintegrationenablestheWED-SQLtotakefulladvantageoftransactionalpropertiesandalsobenefitfromtheSQLlanguagetospecifytheWED-flowdefinitions.
QueryingWorkflowLogsYanTang(UniversityofCaliforniaatSantaBarbara),JianwenSu(UniversityofCaliforniaatSantaBarbara)
Abusinessprocess(BPorworkflow)isanassemblyoftaskstoaccomplishabusinessgoal.Businessprocessmanagement(BPM)isastudytoprovidesupportforthedesign,configuration/implementation,enactmentandmonitoring,diagnose/analysis,andre-designofworkflow.Businessanalyticsorintelligence(BI)isanecessarysteptowardsre-design/improvement.ThetraditionalmethodologyforBIisthewellknownsequenceofETL,data/processwarehouse,andOLAPtools.Inthispaper,wefocusontheproblemofadhocqueryingprocessenactmentsfordata-centricbusinessprocesses.Wedevelopanalgebraicquerylanguagebasedon“incidents”toallowtheusertoformulateadhocqueriesdirectlyonworkflowlogs.Aformalsemanticsandanpreliminaryqueryevaluationalgorithmareprovided.
Ontheintegrationofevent-basedandtransaction-basedarchitecturesforSupplyChainsZhijieLi(IndianaUniversity–PurdueUniversityIndianapolis),HaoyanWu(IndianaUniversity–PurdueUniversityIndianapolis),BrianKing(IndianaUniversity–PurdueUniversityIndianapolis),ZinaBen-Miled(IndianaUniversity–PurdueUniversityIndianapolis),JohnWassick(TheDowChemicalCompany),JeffreyTazelaar(TheDowChemicalCompany)
Affordableandreliablesupplychainvisibilityisbecomingincreasinglyimportantasthecomplexityofthenetworkunderlyingsupplychainsisbecomingordersofmagnitudeshighercomparedtoadecadeago.Moreoverthisincreaseincomplexityisstartingtoreflectonthecostofgoodsandtheiravailabilitytotheconsumers.Thispaperaddressestwokeyissuesinthedistributionphaseofthesupplychain,namely,affordabilityandpseudoreal-timevisibilityoftruckloadactivities.Theproposedframeworkcreatesadigitalthreadthattracksthepseudoreal-timestatusoftheshipmentmakingthephysicaldistributionprocesscompletelytransparenttothestakeholders.Thearchitectureoftheframeworkisbasedonadynamichybridpeer-to-peernetworkandaprivate/publicblockchaindatamodelthatleveragesemergentsensortechnologies.
CacheDOCS:ADynamicKey-ValueObjectCachingServiceJulienGascon-Samson(UniversityofBritishColumbia),MichaelCoppinger(McGillUniversity),FanJin(McGillUniversity),JörgKienzle(McGillUniversity),BettinaKemme(McGillUniversity)
Cachingplaysanimportantroleinmanydomains,asitcanleadtoimportantperformanceimprovements.Akey-valuebasedcachingsystemtypicallystorestheresultsofpopularqueriesinefficientstoragelocation.Whilecachingenjoyswidespreadusageinthecontextofdynamicwebapplications,mostmainstreamcachingsystemsstorestaticbinaryitems,whichmakesthemimpracticalformanyreal-worldapplicationsthatwouldbenefitfromstoringdynamicitems.Inthispaper,weproposeCacheDOCS,adynamickey-valueobjectcachingservicethatallowsforcachingarbitraryobjects.Aspartofourmodel,CacheDOCSprovidesanAPIthatsupportstheexecutionofoperationsagainstcachedobjects,andallowsforclientstoseamlesslysubscribetokeeptheirlocalcopiesinsyncwithcachedremoteobjects.CacheDOCSsupportsmultipleupdatedisseminationstrategiesinordertooptimizeperformance,andproposesaversioningmechanismtoensureconsistency.WeimplementedafullversionofCacheDOCSandweranseveralperformance-relatedexperimentsunderthreeuse-casescenarios.
![Page 135: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/135.jpg)
135
PSBD 2017 Workshop Abstracts
Anovelgame-theoreticmodelforcontent-adaptiveimagesteganographyQiLi(HunanUniversity),XinLiao(HunanUniversity),GuoyongChen(HunanUniversity),LipingDing(GuangzhouBranchofInstituteofSoftware,ChineseAcademyofScience)
Content-adaptiveimagesteganographymeansthatsteganographerchoosessecurityembeddingpositionsbasedonimagetextures.Steganalystcanalsofocusondetectingthesepositionsaccordingtoimagetextures.Gametheoryispreferredtoanalyzetheabovesituation.However,inpreviousgamemodels,steganalystwillmistakenlyidentifythatnobitisembedded,whenthesecretbitisthesameastheleastsignificantbitofcoverimage.Inthispaper,anovelgametheoreticmodelbasedonsecondaryembeddingisproposedtocorrectthejudgmentdrawbackforabetterNashequilibriumbysteganalyst.However,steganalyst’schoicedisturbspreviousequilibriumandsteganographerwillchangehischoicetofindnewequilibriumbyGametheory.Co-occurrencematrixandpointdeviationdegreeareutilizedfordescribingsteganalyst’schoices.Theoccurrencenumberofeachpixelpairsiscalculatedtoconstituteco-occurrencematrix,andthenEuclideandistancebetweenonepointandadjacentpointsiscomputedtolocateembeddingpositions.Incontent-adaptiveimagesteganography,wecandrawaconclusionthatsteganographershouldselectembeddingpositionsfrombothimageedgeareasandsmoothareas.
AFine-grainedAccessControlSchemeforBigDataBasedonClassificationAttributesTengfeiYang(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),PeisongShen(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),XueTian(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences),ChiChen(StateKeyLaboratoryofInformationSecurity,InstituteofInformationEngineering,ChineseAcademyofSciences)
Inordertoprotectthesecurityandprivacyofbigdata,thecloudstorageserviceneedstoenforceeffectiveaccesscontrolmechanismonuserrequests.Attribute-BasedEncryptionisapromisingcryptographicaccesscontroltechniquetoensuretheend-to-endsecurityofdataincloud.However,theexistingABEresearchesmainlyfocusontheefficiencydecryption,whiletheflexibilityofpolicy,thecommunicationcost,andthemetadatamanagementofciphertextsarestillchallengingissuesinthebigdataenvironment.Inthispaper,forthefirsttime,weproposeanewdistributed,scalableandfine-grainedaccesscontrolschemebasedonclassificationattributesforthecloudobjectstorage.Theclassificationattributesandthresholdpoliciesareintegratedintoanaccessstructure,andthentheobjectsareencryptedwiththeintegratedaccessstructure.Theconstant-sizeciphertextcomponentsrelatedtoattributescanbemanagedasthecorrespondingmetadata.Asaresulttheencryptioncomplexityandciphertextstoragearereduced.Inaddition,wepresentanewlabel-basedaccesscontrolmodelwithmulti-authoritiestodescribethedetailedrelationshipsofentitiesinourscheme.Besides,theproposedschemeisprovedtobesecureunderlBDHEassumption,andthesystemimplementationdemonstratesthepracticalfeasibilityandgoodperformance.
Social-AwareDecentralizationforEfficientandSecureMulti-PartyComputationYuzheTang(SyracuseUniversity),SuchetaSoundarajan(SyracuseUniversity)
ThisworkstudiestheproblemofMPCschedulingthatis,identifyingasetofcomputingnodestoexecutesecuremulti-partycomputationprotocols(MPC)overadistributedprivatedataset.Ourprimarycontributionisinestimatingtheriskofcollusionbetweennodestowhomthecomputationisscheduled.Thisworkhaspotentialinenablingefficientprivacy-preservingdatasharinginemergingplatformsofbig-datafederation,inhealthcare,finance,andothermarketplaces.Inourmethods,weassumethattheMPCcomputingnodesexistinasocialnetwork,andpresenttwomodelsforestimatingtheriskofcollusion,aswellasalgorithmsforfindingtheMPCnodessuchthattheriskofcollusionisminimized.Weevaluateourmethodsonseveralreal-worldnetworkdatasets,andshowthattheyareeffectiveinminimizingtherisklevels.
StatisticalAnomalyDetectiononMetadataStreamsviaCommoditySoftwaretoProtectCompanyChristineChen(UniversityofPortland),JamesGurganus(MicroSystemsEngineering,Inc.)
Asacompanygrows,itsinfrastructurenaturallymustgrowtosupportit.Theresultingmountainsofinfrastructuremetadatacontainvaluableinformationonthehealthandwellbeingofthesystemsthroughoutthecompany.Forexample,anabnormallylowdiskwriteratetoafileservermayindicatethataregularlyscheduledtaskhasfailedtostart,oranabnormallyhighdiskwriteratemayindicatethepresenceofamaliciousthreatsuchasransomware.Thehypothesisofthiscasestudyisthatsuchmetadatastreamscanbeeffectivelyutilizedbyimplementingstatisticalanomalydetectionmethodsviacommoditysoftware(Splunk,inthiscase).Thesemethodsweretestedprimarilyonservermetadatainaransomwaresimulationandalsoonservermetadatafromfileserversandproductionserversinactiveuse.
Intheransomwaresimulation,thealertingsystemdetectedtheransomwarebehaviorfiveminutesafteranencryptioneventbeganinthesimulationenvironmentandalertedsteadilyforthedurationofthesimulation.Intheweek-longexperimentover11fileserversandproductionservers,atotalof1,484alertsweregenerated.Applyingsimplecorrelationtechniquescreatedamoreconcentratedinformationstreamwith77events.Theseresultsconfirmthevalueofmetadatainidentifyingsystemanomaliesandprovidinganotherlayerofdefenseagainstmaliciousthreats.Therelativelysimpleanomalydetectiontechniquesutilizedinthiscasestudyalsohighlighttheincreasingpracticalityofbehavioralanalytics—itcanonlybeamatteroftimebeforesuchtechniqueswillbeubiquitous.
Computationalimprovementsinparallelizedk-anonymousmicroaggregationoflargedatabasesAhmadMohamadMezher(UniversitatPolitècnicadeCatalunya),AlejandroGarcíaÁlvarez(UniversitatPolitècnicadeCatalunya),DavidRebollo-Monedero(UniversitatPolitècnicadeCatalunya),JordiForné(UniversitatPolitècnicadeCatalunya)
Thetechnicalcontentsofthispaperfallwithinthefieldofstatisticaldisclosurecontrol(SDC),whichconcernsthepostprocessingofthedemographicportionofthestatisticalresultsofsurveyscontainingsensitivepersonalinformation,inordertoeffectivelysafeguardtheanonymityoftheparticipatingrespondents.Theconcretepurposeofthisstudyistoimprovetheefficiencyofawidelyusedalgorithmfork-anonymousmicroaggregation,knownasmaximumdistancetoaveragevector(MDAV),tovastlyaccelerateitsexecutionwithoutaffectingitsexcellentfunctionalperformancewithrespecttocompetingmethods.Theimprovementsputforthinthispaperencompassalgebraicmodificationsandtheuseofthebasiclinearalgebrasubprograms(BLAS)library,fortheefficientparallelcomputationofMDAVonCPU.
![Page 136: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/136.jpg)
136
WoSC 2017 Workshop Abstracts
Ripple:HomeAutomationforResearchDataManagementRyanChard(ArgonneNationalLaboratory),KyleChard(UniversityofChicagoandArgonneNationalLab),JasonAlt(NationalCenterforSupercomputingApplications),DilworthParkinson(LawrenceBerkeleyNationalLaboratory),SteveTuecke(UniversityofChicagoandArgonneNationalLab),IanFoster(ArgonneNationalLaboratory&TheUniversityofChicago)
Explodingdatavolumesandacquisitionrates,plusevermorecomplexresearchprocesses,placesignificantstrainonresearchdatamanagementprocesses.Itisincreasinglycommonfordatatoflowthroughpipelinescomprisedofdozensofdif-ferentmanagement,organization,andanalysisstepsdistributedacrossmultipleinstitutionsandstoragesystems.Toalleviatetheresultingcomplexity,weproposeahomeautomationapproachtomanagingdatathroughoutitslifecycle,inwhichusersspecifyviahigh-levelrulestheactionsthatshouldbeperformedondataatdifferenttimesandlocations.Tothisend,wehavedevelopedRIPPLE,aresponsivestoragearchitecturethatallowsuserstoexpressdatamanagementtasksviaarulesnotation.RIPPLEmonitorsstoragesystemsforevents,evaluatesrules,andusesserverlesscomputingtechniquestoexecuteactionsinresponsetotheseevents.WeevaluateoursolutionbyapplyingRIPPLEtothedatalifecyclesoftworeal-worldprojects,inastronomyandlightsourcescience,andshowthatitcanautomatemanymundaneandcumbersomedatamanagementprocesses.
Pipsqueak:LeanLambdaswithLargeLibrariesEdwardOakes(UniversityofWisconsin-Madison),LeonYang(UniversityofWisconsin-Madison),KevinHouck(UniversityofWisconsin-Madison),TylerHarter(MicrosoftGraySystemsLab),AndreaC.Arpaci-Dusseau(UniversityofWisconsin-Madison),RemziH.Arpaci-Dusseau(UniversityofWisconsin-Madison)
Microservicesareusuallyfasttodeploybecauseeachmicroserviceissmall,andthuseachcanbeinstalledandstartedquickly.Unfortunately,leanmicroservicesthatdependonlargelibrarieswillstartslowlyandharmelasticity.Inthispaper,weexplorethechallengesofleanmicroservicesthatrelyonlargelibrariesinthecontextofPythonpackagesandtheOpenLambdaserverlesscomputingplatform.WeanalyzethepackagetypesandcompressibilityoflibrariesdistributedviathePythonPackageIndexandproposePipBench,anewtoolforevaluatingpackagesupport.WealsoproposePipsqueak,apackage-awarecomputeplatformbasedonOpenLambda.
LeveragingtheServerlessArchitectureforSecuringLinuxContainersNiltonBila(IBM),PaoloDettori(IBM),AliKanso(IBM),YujiWatanabe(IBM),AlaaYoussef(IBM)
Linuxcontainerspresentalightweightsolutiontopackageapplicationsintoimagesandinstantiatetheminisolatedenvironments.Suchimagesmayincludevulnerabilitiesthatcanbeexploitedatruntime.Avulnerabilityscanningservicecandetectthesevulnerabilitiesbyperiodicallyscanningthecontainersandtheirimagesforpotentialthreats.Whenathreatisdetected,aneventmaybegeneratedto(1)quarantineorremovethecompromisedcontainer(s)andoptionally(2)remedythevulnerabilitybyrebuildingasecureimage.Webelievethatsuchevent-drivenprocessisagreatfittobeimplementedinaserverlessarchitecture.InthispaperwepresentourdesignandimplementationofaserverlesssecurityanalyticsservicebasedonOpenWhiskandKubernetes.
ServerlessComputing:Design,Implementation,andPerformanceGarrettMcGrath(UniversityofNotreDame),PaulR.Brenner(UniversityofNotreDame)
Wepresentthedesignofanovelperformance-orientedserverlesscomputingplatformimplementedin.NET,deployedinMicrosoftAzure,andutilizingWindowscontainersasfunctionexecutionenvironments.Implementationchallengessuchasfunctionscalingandcontainerdiscovery,lifecycle,andreusearediscussedindetail.WeproposemetricstoevaluatetheexecutionperformanceofserverlessplatformsandconducttestsonourprototypeaswellasAWSLambda,AzureFunctionsandIBM’sdeploymentofApacheOpenWhisk.Ourmeasurementsshowtheprototypeachievinggreaterthroughputthanotherplatformsatmostconcurrencylevels,andweexaminethescalingandinstanceexpirationtrendsintheimplementations.Additionally,wediscussthegapsandlimitationsinourcurrentdesign,proposepossiblesolutions,andhighlightfutureresearch.
![Page 137: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/137.jpg)
137
NSF-JST 2017 Workshop Abstracts
AcceleratingBigDataInfrastructureandApplicationsKevinBrown(TokyoInstituteofTechnology),TianqiXu(TokyoInstituteofTechnology),KeitaIwabuchi(TokyoInstituteofTechnology),KentoSato(LawrenceLivermoreNationalLaboratory),AdamMoody(LawrenceLivermoreNationalLaboratory),KathrynMohror(LawrenceLivermoreNationalLaboratory),NikhilJain(LawrenceLivermoreNationalLaboratory),AbhinavBhatele(LawrenceLivermoreNationalLaboratory),MartinSchulz(LawrenceLivermoreNationalLaboratory),RogerPearce(LawrenceLivermoreNationalLaboratory),MayaGokhale(LawrenceLivermoreNationalLaboratory),SatoshiMatsuoka(TokyoInstituteofTechnology)
High-performancecomputing(HPC)systemsareincreasinglybeingusedfordata-intensive,or``BigData",workloads.However,sincetraditionalHPCworkloadsarecompute-intensive,theHPC-BigDataconvergencehascreatedmanychallengeswithoptimizingdatamovementandprocessingonmodernsupercomputers.Ourcollaborativeworkaddressesthesechallengesusingathree-prongedapproach:(i)measuringandmodelingextreme-scaleI/Oworkloads,(ii)designingalow-latency,scalable,on-demandburst-buffersolution,and(iii)optimizinggraphalgorithmsforprocessingBigDataworkloads.Wedescribethethreeareasofourcollaborationandreportontheirrespectivedevelopments.
DisasterNetworkEvolutionUsingDynamicClusteringofTwitterDataKrishnaKant(TempleUniversity),YilangWu(AizuUniversity),ShanshanZhang(TempleUniversity),JunboWang(AizuUniversity),AmitangshuPal(TempleUniversity)
Adhocsmartphonenetworkscanbeusedtoaugmentcommunicationsdegradedbydisastersprovidedthattheindividualadhocclusterscanreachsome``connectiongateways''togetouttotheInternetviaconnecteddevicesinthesurroundingarea(inadditiontoconnectivityviaanyspeciallydeployedemergencyequipment).Thedisconnectedareasarenotknownuntiltheyarebackonline;however,weneedamechanismtodeterminethemsothatthegatewaydevicecanbebestrecruitedtoprovidetheconnectivity.Thisneedstobedoneinadynamicenvironmentbecauseofdisasterrelatedmobility.Inthispaperweproposeamechanismtosolvethisproblembyestimatingregionsthatarelikelytobedensebutdisconnectedwithsignificantnumberofconnecteddevicesaroundthem.Becauseoflackofdirectinformationonpeople(orsmartphone)density,weattempttodothisbyanalyzingthetwitterdata.Byvirtueofitsefficiency,thealgorithmcanbeusedonadynamicallyevolvingdatasetandthusallowsdynamictracking.
Single-epochsupernovaclassificationwithdeepconvolutionalneuralnetworksAkisatoKimura(NTT),IchiroTakahashi(KavliIPMU,TheUniversityofTokyo),MasaomiTanaka(NationalAstronomicalObservatoryofJapan),NaokiYasuda(KavliIPMU,TheUniversityofTokyo),NaonoriUeda(NTT),NaokiYoshida(KavliIPMU,TheUniversityofTokyo)
SupernovaeType-Ia(SNeIa)playasignificantroleinexploringthehistoryoftheexpansionoftheUniverse,sincetheyarethebest-knownstandardcandleswithwhichwecanaccuratelymeasurethedistancetotheobjects.FindinglargesamplesofSNeIaandinvestigatingtheirdetailedcharacteristicshasbecomeanimportantissueincosmologyandastronomy.Thecurrentphotometricsupernovasurveysproducevastlymorecandidatesthancanbefollowedupspectroscopically,highlightingtheneedforeffectiveclassificationmethods.Existingmethodsreliedonaphotometricapproachthatfirstmeasurestheluminanceofsupernovacandidatespreciselyandthenfitstheresultstoaparametricfunctionoftemporalchangesinluminance.However,itinevitablyrequiresalotofobservationsandcomplexluminancemeasurements.Inthiswork,wepresentanovelmethodfordetectingSNeIasimplyfromsingle-shotobservationimageswithoutanycomplexmeasurements,byeffectivelyintegratingthestate-of-the-artcomputervisionmethodologyintothestandardphotometricapproach.Ourmethodfirstbuildsaconvolutionalneuralnetworkforestimatingtheluminanceofsupernovaefromtelescopeimages,andthenconstructsanotherneuralnetworkfortheclassification,wheretheestimatedluminancesandobservationdatesareusedasfeaturesforclassification.BothoftheneuralnetworksareintegratedintoasingledeepneuralnetworktoclassifySNeIadirectlyfromobservationimages.Experimentalresultsshowtheeffectivenessoftheproposedmethodandrevealclassificationperformancecomparabletoexistingphotometricmethodswithmanyobservations.
EnablingLargeScaleDeliberationusingIdeationandNegotiation-SupportAgentsKatsuhideFujita(TokyoUniversityofAgricultureandTechnology),TakayukiIto(NagoyaInstituteofTechnology),MarkKlein(MIT)
ThispaperdescribesanongoingJapan-USprojectthatisdevelopingthekindofadvancedcomputersupportforonlinecrowd-scaledeliberationthatisneededtoenablesmarterandmoreconnectedcommunities.Oursharedworkhasfocusedonaddressingboththeseproblems:(1)ideation:helpingcrowdsmoreeffectivelydeveloppotentialwin-winsolutions,and(2)decision-making:helpingcrowdsgettopareto-optimalityinthesolutionstheyselect.InJapan,adiscussionsupportsystemcalledCOLLAGREEthatfacilitatesfreetextdiscussionstoachieveconsensushasbeendeveloping.InUS,anonlinetoolcalledtheDeliberatoriumthatintegratesargumentationtheoryandsocialcomputingtechniquestoenablemoreeffectivecrowd-scaledeliberationhasbeendeveloping.Oneofourimmediatejointworkistointegratethefacilitatedfree-textdiscussionsofCOLLAGREEwiththestructureddeliberationsprovidedbytheDeliberatorium.Wewillalsodevelopautomatedagentsthatenablebetterideationaswellasbetterdecision-making.
![Page 138: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/138.jpg)
138
Local Information
Maps
![Page 139: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/139.jpg)
139
Restaurants in/near Lenox Mall
![Page 140: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/140.jpg)
140
![Page 141: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/141.jpg)
141
![Page 142: ICDCS 2017: Program at a Glanceicdcs2017.gatech.edu/.../ICDCS2017-program-posting-v17.pdfVision/Blue Sky Thinking Track Paper Abstracts ..... 88 Short Paper Abstracts ..... 95 Demonstration](https://reader031.vdocuments.net/reader031/viewer/2022020718/5ae0dcdb7f8b9a8f298eb6cd/html5/thumbnails/142.jpg)
142