cloud futurology - posco, fung po tso · cloud futurology blesson varghese 1, philipp leitner2,...
TRANSCRIPT
Cloud Futurology
Blesson Varghese∗1, Philipp Leitner2, Suprio Ray3, Kyle Chard4,Adam Barker5, Yehia Elkhatib6, Herry Herry7, Cheol-Ho Hong8,
Jeremy Singer9, Fung Po Tso10, Eiko Yoneki11, and Mohamed-Faten Zhani12
1Queen’s University Belfast, UK; [email protected] University of Technology, Sweden; [email protected]
3University of New Brunswick, Canada; [email protected] of Chicago, USA; [email protected]
5University of St Andrews, UK; [email protected] University, UK; [email protected]
7University of Glasgow, UK; [email protected] University, South Korea; [email protected]
9University of Glasgow, UK; [email protected] University, UK; [email protected]
11University of Cambridge, UK; [email protected] de Technologie Superieure, Canada; [email protected]
The Cloud has become integral to most Internet-based applications and user gadgets. Thisarticle provides a brief history of the Cloud and presents a researcher’s view of the prospectsfor innovating at the infrastructure, middleware, and application and delivery levels of thealready crowded Cloud computing stack.
The global Cloud computing market exceeds$100 billion and research in this area hasrapidly matured over the last decade. Dur-
ing this time many buzz words have come and gone.Consequently, traditional concepts and conventionaldefinitions related to the Cloud are almost obso-lete [1]. The current Cloud landscape looks very dif-ferent from what may have been envisioned duringits inception. It may seem that the area is saturatedand there is little innovative research and develop-
ment to be done in the Cloud, which naturally raisesthe question - ‘What is the future of the Cloud?’
This article first examines the multiple genera-tions of innovation that the Cloud has undergone inthe last decade, it then presents a researcher’s viewof the prospects and opportunities for innovation inthis area across the entire Cloud stack - highlightingthe infrastructure, middleware, and application anddelivery levels.
∗Corresponding Author
1
Fig
ure
1:A
tim
elin
eof
the
Clo
ud
com
pu
tin
gla
nd
scap
e.E
arly
men
tion
sof
the
Clo
ud
inli
tera
ture
are
des
crib
edin
the
gre
enb
lock
.T
he
per
iod
ofin
cep
tion
ish
igh
ligh
ted
intu
rqu
oise
.T
his
isfo
llow
edby
two
gen
erati
ons
of
dev
elop
men
tson
the
Clo
ud
-th
eora
nge
blo
ckh
igh
ligh
tsfi
rst
gen
erat
ion
and
the
pu
rple
colo
ur
pre
sents
seco
nd
gen
erati
on
dev
elop
men
ts.
Up
com
ing
dev
elop
men
tsan
dtr
end
sar
ep
rese
nte
das
nex
tge
ner
atio
nd
evel
opm
ents
inth
eyel
low
blo
ck.
2
The Cloud Landscape
The first known suggestion of Cloud-like com-puting was by Professor John McCarthy atMIT’s centennial celebration in 1961 “Com-puting may someday be organized as a publicutility just as the telephone system is a publicutility ... Each subscriber needs to pay only forthe capacity he actually uses, but he has accessto all programming languages characteristic ofa very large system ... Certain subscribersmight offer service to other subscribers ... Thecomputer utility could become the basis of anew and important industry”
Two fundamental technologies required for devel-oping the Cloud, namely virtualization and network-ing, were first developed in the 60’s. In 1967, IBMvirtualized operating systems to allow multiple usersto share the same computer and in 1969, the US De-partment of Defense launched the Advanced ResearchProjects Agency Network (ARPANET), which de-fined network protocols that led to the developmentof the Internet. Although the earliest mentions ofCloud computing in literature appear in the 90s asshown in Figure 1, it was Grid computing [2] thatlaid the foundation for offering computing resourcesas a service to users in the 90’s and early 21st cen-tury. The inception of the Cloud as a utility servicewas realized when Amazon launched its commercialpublic Cloud in 2002.
Significant advances over the last decade can bedivided into two generations as seen in Figure 1.The first generation focuses on development at theinfrastructure level, for example creating data cen-tres, which are centralized infrastructure that hostsignificant processing and storage resources acrossdifferent geographic regions. Across other layersof the Cloud stack a range of user-facing servicesemerged, some of which were available only in specificgeographic regions. Software developed by Open-Nebula (http://www.opennebula.org) and Open-Stack (https://www.openstack.org) allowed orga-nizations to own private Clouds and set up their owndata centres.
As big data started to gain popularity in 2005,the Cloud was a natural first choice to tackle big
data challenges. This led to the popularity of storageservices such as Dropbox that relied on the Cloud.Relational databases hosted in the Cloud to supportenterprise applications emerged.
Since inception in early 2000 and despite sig-nificant research and development efforts span-ning over half a decade, a reference archi-tecture for the Cloud was not defined un-til 2011 (https://ws680.nist.gov/publication/get_pdf.cfm?pub_id=909505). Since the Cloud wasa new technology it may have taken a few years beforedefinitions were articulated, circulated, and widelyaccepted.
Second generation developments focused on en-riching the variety of services and their quality. Inparticular, management services and modular appli-cations emerged. Monitoring services of compute,network and storage resources offering aggregate andfine metrics became available to application owners,allowing them to maximize performance. More flex-ible pricing strategies and service level agreements(SLAs), in addition to the posted price, pay-as-you-go model, such as spot bidding and preemptible vir-tual machine instances emerged in 2010.
Furthermore, the move from immutable virtualmachines to smaller, loosely coupled execution unitsin the form of microservices and containers was agame changer for decomposing applications withinand across different data centres. Combining pub-lic and private (on-premise) Clouds of different scale(a.k.a cross-cloud or hybrid Cloud computing [3])gained prominence in order to alleviate concerns re-lated to privacy and vendor lock-in.
An important step in the evolution of the Cloudwas the development of Content Delivery Networks(CDNs). Compute and storage resources were geo-graphically distributed for improving the overall qual-ity of a variety of services, including streaming andcaching. CDNs are the basis of upcoming trends indecentralizing Cloud resources towards the edge ofthe network, which will be considered in this article.Amazon launched their CDN in 2008.
3
Containers are namespaces with ring-fencedresources. For example, Docker [4] is a popu-lar container technology for creating and man-aging self-sufficient execution units. Contain-ers offer the prospect of seamless applicationdevelopment, testing, and delivery over het-erogeneous environments. The microservicearchitectural style focuses on how the appli-cation logic is implemented rather than howit is hosted by dividing services into atomicfunctions in order to tame operational com-plexity. The emphasis is on developing small,replaceable service units instead of maintain-ing monolith services [5].
Innovation for the Next
Generation
Tangential innovations in the first and second gener-ation developments have made way for another gen-eration of Cloud development that will focus on de-centralization of resources. Although there has beena decade long explosion of Cloud research and devel-opment, there is significant innovation yet to come inthe infrastructure, middleware, and application anddelivery areas.
As shown in the ‘Next Generation Development’block of Figure 1 in the next five years, computingas a utility will be miniaturized and available outsidelarge data centres. Referred to as ’Cloud-in-a-Box’,Sun Microsystems first demonstrated these ideas in2006 and paved the way for Fog/Edge computing.The use of hardware accelerators, such as Graph-ics Processing Units (GPUs), in the Cloud began in2010 is now leading to inclusion of even more special-ized accelerators that are, for example, customizedfor modern machine learning or artificial intelligenceworkloads. Google provides Tensor Processing Units(TPUs) that are customized for such workloads withthe aim to deliver new hardware and software stacksthat extend machine learning and artificial intelli-gence capabilities both within and outside the cloud.
Infrastructure
A range of hardware accelerators, such as GraphicsProcessing Units (GPUs), Field-Programmable GateArrays (FPGAs), and more specialized Tensor Pro-cessing Units (TPUs) are now available for improvingthe performance of applications on the Cloud. Typ-ically, these accelerators are not shared between ap-plications and therefore result in an expensive datacentre setup given the large amount of underutilizedhardware.
Accelerator virtualization is the underlyingtechnology that allows multiple applications to sharethe same hardware accelerator [6]. All existing virtu-alization solutions have performance limitations andare bespoke to each type of accelerator. Given thatthis is a relatively new area of study, we are yet tosee a robust production-level solution that can easilyincorporate and virtualize different types of accelera-tors with minimal overheads.
Currently, applications leverage Cloud resourcesfrom geographically distant data centres. Conse-quently, a user device, such as a smartphone musttransfer data to a remote Cloud for processing data.However, this will simply not be possible when bil-lions of devices are connected to the Internet, as fre-quent communication and communication latencieswill affect the overall service quality and experienceof the user. These latencies can be reduced by bring-ing compute resources to the network edge in a modeloften referred to as Fog/Edge computing [7, 8].
Edge computing is challenging - Edge resourcesneed to be publicly and securely available. However,the risks, security concerns, vulnerabilities and pric-ing models are not articulated, or even fully known.Implementations of standardized Edge architecturesand a unified marketplace is likely to emerge. Fur-thermore, there are no robust solutions to deploy anapplication across the Cloud and the Edge. Toolk-its for deploying and managing applications on theEdge will materialize given the community led effortsby the European Telecommunications Standards In-stitute, the OpenFog consortium, and OpenStack.
4
Fog/Edge computing [7,8] refers to the use ofresources at the Cloud to the network edgecontinuum. Data from user devices can beprocessed at the edge instead of remote datacenters. Gartner and Forester define Edgecomputing as an enabling and strategic tech-nology for realizing IoT and for building fu-ture Cloud applications. The Edge comput-ing market is estimated to be worth US $6.72billion within the next five years with a com-pound annual growth rate of over 35%.
The definition of Cloud data centres is also chang-ing - what was conventionally ‘dozens of data centreswith millions of cores,’ is now moving towards ‘mil-lions of data centres with dozens of cores.’ Thesemicro data centres are novelty architectures andminiatures of commercial cloud infrastructure (seeFigure 2). Such deployments have become possibleby networking several single board computers, suchas the popular Raspberry Pi [9]. Consequently, theoverall capital cost, physical footprint and power con-sumption is only a small fraction when compared tothe commercial macro data centres [10].
Figure 2: A micro data centre comprising a cluster ofRaspberry Pis that was assembled at the Universityof Glasgow, UK. These data centres in contrast tolarge Cloud data centres are low cost and low powerconsuming.
Micro data centres are a compelling infras-tructure to support education, where studentscan be exposed to near-real data centre sce-narios in a sandbox. They are widely usedin educational contexts and more convincinguse-cases will emerge. Global community en-gagement will facilitate the further adoptionof micro data centres. International competi-tions and league tables for micro data centredeployment would catalyze this process.
The challenges to be addressed in making microdata centres operational include reducing overheadsof the software infrastructure. For example, machinelearning libraries must be optimized for use on mi-cro data centre hardware. Additionally, managementtools must run on devices that have only limited net-work access, perhaps due to the presence of firewalls,or intermittent connectivity as commonly found onedge networks. This is different from traditional datacentres, which operate on the assumption that man-aged nodes are directly and permanently reachable.
Data centers are now one of the largest consumersof energy (http://www.climatechangenews.com/2017/12/11/tsunami-data-consume-one-fifth-
global-electricity-2025/). This is in part due tothe end of Moore’s law and the fact that increasingprocessor speed no longer offsets energy consump-tion. It is also due to the rapid growth of Internet-of-Things (IoT) - with estimates of up to 80 billiondevices online by 2025 - and increasing use in devel-oping countries. As such, Cloud providers are facingboth economic and legislative pressures to reduceenergy consumption. It is unlikely that new powerplants will be sufficient to meet these growing en-ergy needs and thus there will be widespread use ofrenewable energy sources and ‘stranded power’ [11].
Middleware
Cloud customers are spoiled for choice, to the extentthat it has become overwhelming for many. This isbecause of the incredible rate at which the Cloud re-source and service portfolio has expanded. As such,there is now a need not only for middleware tech-nologies that abstract differences between Clouds andservices, but also for decision support systems to aid
5
customer deployment of applications. For example,to help guide selection of the best Cloud providers,resources and services, and configuration.
Cross-cloud challenges, such as identifying opti-mal deployment resources from across the vast arrayof options from different providers, seamlessly movingworkloads between providers, and building systemsthat work equally well across the services of differentproviders will need to be surmounted for establishinga viable Cloud federation .
Such systems, naturally, cannot be one-size-fits-all, but they must be tailored to the needs of the cus-tomer and to follow any changes in the Cloud provi-sioning market. Resource brokers are likely to be-come necessary to fill this void and provide a means ofexploiting differences between Cloud providers, andidentifying the real performance profiles of differentCloud services before matching them to the customerneeds. Currently however, no practical brokering so-lutions are available.
Brokers can also be used to automatically config-ure the parameters of applications for a set of selectedresources so as to maximize the performance of ap-plications. Typically, the configuration parametersare manually tuned, which is cumbersome given theplethora of resources. A generic auto-tuner thatoperates in near real-time and measures performancecost-effectively is ideal. However, measuring perfor-mance in the Cloud is time consuming and thereforeexpensive [12,13]. In a transient environment, such asthe Cloud, the performance metrics will be obsolete ifthey cannot be gathered in real time. The complex-ity increases when resources from multiple providersare considered. A starting point may be to developbespoke auto-tuners based on domain specific knowl-edge that a system engineer may possess using a com-bination of machine learning techniques, such as deepneural networks and reinforcement learning [14].
While resource brokerage allows customers to se-lect and combine services to their liking, Cloud datacentre operators can also choose a variety of NetworkFunctions (NFs) to create in-network services andsafeguard networks to improve an application’s per-formance. The process of orderly combining differentcombination or subsets of NFs, such as firewalls, net-work monitors, and load balancers, is referred to asservice chaining [15].
Service chaining will be particularly useful for
Edge computing to, for instance, improve data pri-vacy. Recent proposals of intent-driven network-ing [16] allow operators and end-users to define whatthey want from the network, not how. This enablesthe on demand composition of bespoke network logic,allowing much more refined application control anddynamicity. A research challenge here is to man-age services across Cloud and Edge networks and re-sources that have different ownership and operatingobjectives.
An interesting starting point for implementingservice chaining will be creating personalizednetwork services across Cloud and Edge envi-ronments. This will provide, for example, auser with a personalized security profile thatis adaptive to different environments, such ashome or office. Network functions will needto be miniaturized for Edge resources to facil-itate chaining.
Application and Delivery
The Cloud will continue to be an attractive proposi-tion for big data applications to meet the volume,velocity, and variety challenges [17]. Apache Spark,Hadoop MapReduce and its variants have been ex-tensively used to process volumes of data in the lastfive years. However, for many users these frameworksremain inaccessible due to their steep learning curveand lack of interactivity.
Further, while frameworks such as Storm andKafka address some of the challenges associated withdata velocity (e.g., as seen in real-time streaming re-quired by social media, IoT and high frequency trad-ing applications), scaling resources on the Cloud tomeet strict response time requirements is still an ac-tive research area. This requires complex stream pro-cessing with low latency, scalability with self-load bal-ancing capabilities and high availability. Many ap-plications that stream data may have intermittentconnectivity to Cloud back-end services for data pro-cessing and it would be impossible to process all dataat the edge of the network. Thus, efficient streamprocessing frameworks that can replicate stream op-erators over multiple nodes and dynamically routestream data to increase potential paths for data ex-
6
change and communication will be desirable.Innovation will be seen in taming traditional data
challenges. For example, emergence of tools that al-leviate the burden on the user, that support elasticity- dynamic provisioning and fair load balancing, andthat cleverly move data. As data are increasinglylarge and distributed, the cost of moving data cannow exceed the cost of processing it. Thus, there willbe increasing interest in moving computation to dataand constructing federated registries to manage dataacross the Cloud.
A new class of distributed data management sys-tems, called NewSQL [18], are emerging. These sys-tems aim to offer similar scalable performance toNoSQL while supporting Atomicity, Consistency, Iso-lation, and Durability (ACID) properties for transac-tional workloads that are typical of traditional rela-tional databases. However, providing ACID guaran-tees across database instances that are distributedacross multiple data centers is challenging. This issimply because it is not easy to keep data replicasconsistent in multiple data locations. More matureapproaches that will allow for data consistency willemerge to support future modularization of applica-tions on the Cloud.
A variety of existing and upcoming applications,such as smart homes, autonomous trading algorithmsand self-driving cars, are expected to generate andprocess vast amounts of time-series data. These ap-plications make decisions based on inputs that changerapidly over time. Conventional database solutionsare not designed for dealing with scale and easy-useof time-series data. Therefore, time-series databasesare expected to become more common.
Gartner predicts that by 2022 nearly 50% ofenterprise-generated data will be processedoutside the traditional Cloud data centre.Processing data outside the Cloud is thepremise of Edge computing. It is likely thatnovel methods and tools that perform complexevent processing across the Cloud and Edgewill emerge.
A new class of applications are starting to emergeon the Cloud, namely serverless applications.They are exemplified by the Function-as-a-Service(FaaS) Cloud, such as AWS Lambda or Azure Cloud
Functions. FaaS Clouds are implemented on topof upcoming containerization technology, but provideconvenient developer abstractions on top of, for in-stance, Docker. Contrary to traditional Cloud appli-cations that are billed for the complete hour or theminute, serverless applications are billed by the mil-lisecond [19]. FaaS has not yet seen widespread adop-tion for business-critical Cloud applications. This isbecause FaaS services and tooling are still immatureand sometimes unreliable Furthermore, since FaaSClouds rely on containers significant overheads areincurred for on-demand boot up. This may be prob-lematic for an end-user facing use cases. FaaS Cloudshave limited support for reuse, abstraction and mod-ularization, which are usually taken for granted whenusing distributed programming models. Anotherpractical challenge is that current-day FaaS servicesare entirely stateless. All application state needs tobe handled in external storage services, such as Redis.FaaS providers are currently working towards statefulstorage solutions, but it is as of yet unclear what thesesolutions will look like in practice. Consequently, sig-nificant developer effort is currently required to takeadvantage of FaaS services.
FaaS Clouds have obvious advantages and wewill witness innovation at the virtualizationfront either to reduce the overheads of contain-ers or in the development of entirely new light-weight virtualization technologies. More pow-erful programming abstractions for composingand reusing Cloud functions will emerge.
Delivery of cloud services via economic modelsis transforming computing into a commodity, inlinewith McCarthy’s vision of computing utility. Theseeconomic models, and benefits from economies ofscale, have underpinned much of the success of theCloud. The Cloud now utilizes a range of economicmodels, from posted price models, through to dy-namic, spot markets for delivering infrastructure re-sources and a suite of higher-level platform and soft-ware services [20]. AWS even allows users to directlyexchange reserved resources with one another via areseller market. As the Cloud moves towards furtherdecoupled resources (e.g., as seen in serverless com-puting) new economic models are needed to addressgranular resource bundles, short-duration leases, and
7
flexible markets for balancing supply and demand.Furthermore, granular, differentiated service levelsare likely to become common, enabling greater userflexability with respect to price and service quality.This increasingly diverse range of economic modelswill further enable flexibility; however, it will requiregreater expertise to understand trade-offs and effec-tively participate in the market. Cloud automationtools will emerge to alleviate this burden by enablingusers to directly quantify and manage inherent trade-offs such as cost, execution time, and solution accu-racy.
Current Cloud providers operate as indepen-dent silos, with little to no ability for usersto move resources between providers. Froma technology standpoint, creating federatedClouds via standardized abstractions is a so-lution, but general markets in which Cloudofferings can be compared and delivered areexpected to appear soon. Users may be po-tentially offered many interchangeable alter-natives, and therefore new economic modelswill need to cater for competition betweenproviders rather than consumers.
Opportunities and Outlook
This article highlights a researcher’s view of theprospects at the infrastructure, middleware, and ap-plication and delivery level of the Cloud computinglandscape.
Although there are ongoing efforts to tackle re-search challenges at the infrastructure level in fouravenues, namely accelerator virtualization, Fog/Edgecomputing, micro data centres, and power andenergy-aware solutions, the timeline to mass adoptionwill vary. rCUDA (http://rcuda.net/) and gVirtus(https://github.com/RapidProjectH2020/GVirtuS)are exemplars of accelerator virtualization solutionsthat have undergone a decade of innovation, butare yet to become available on production levelCloud systems. With the increasing number of ac-celerators used in the Cloud it is foreseeable thataccelerator virtualization technologies are adoptedwithin the next decade. Similarly, there are lim-
ited Fog/Edge computing test-beds and real deploy-ments of such systems are yet to be seen. Whilethe mechanisms required to adopt Fog/Edge sys-tems are currently unknown, the growing empha-sis on 5G communication will accelerate interest inthe area in the short term. Most micro data cen-tres are still prototypes (https://news.microsoft.com/features/under-the-sea-microsoft-tests-
a-datacenter-thats-quick-to-deploy-could-
provide-internet-connectivity-for-years/)and will become widespread as more compelling use-cases, for example in Fog/Edge computing emergeover the next five years.
At the middleware level Cloud federation, re-source brokers and auto-tuners, and service chain-ing were considered. Cloud federation in its strictestsense, i.e. reaching operational arrangement betweenindependent providers, has for long been discussedbut never in fact emerged on the horizon. Never-theless, efforts into tools to bridge services of dif-ferent providers are expected to continue. Currentresource brokers and auto-tuners focus on the chal-lenge of abstraction, but very few tackle this alongwith delegation, i.e., fully adaptive life cycle man-agement. With increasing use of machine learning,the mechanisms of enabling such smart brokerageand auto-tuning are becoming available. The relatedchallenges of expressing and interpreting customer re-quirements is also becoming a significant trend to-wards a vision where the customer needs to know farless about the infrastructure than current DevOps.In a similar vein, capturing and satisfying what anapplication needs from the network as seen in servicechaining is a prominent research challenge. However,network operators are not as advanced in the pay-as-you-go model as cloud providers are. We expect thisto change over the next five years, opening up a widerange of resources for supporting applications acrossan end-to-end network.
The four key areas identified for innovation at theapplications and delivery level are complex streamprocessing, big data applications and databases,serverless applications and economic models. Therapid increase in connected devices, IoT, and stream-ing data is driving the immediate development of newreal-time stream processing capabilities. In the nextfive years, we expect to see a variety of new streamprocessing techniques, including those designed to
8
leverage edge devices. Big data applications anddatabases have been the focus of recent research andinnovation, and thus much immediate focus is onthe adoption and use of these efforts. In the next2-5 years, there will be increasing effort focused ontime-series databases (following on from the releaseof Amazon Timestream and Azure Time Series In-sights). Serverless computing is still in its infancyand is likely to be the most disruptive of the changesat the applications and delivery level. Over the nextten years, serverless technologies will likely perme-ate IT infrastructure, driving enhancements in termsof performance, flexibility (for example, support formore general application types), and economic mod-els. Economic models will evolve over the next 2-5years due to maturing serverless and decoupled com-puting infrastructure. The underlying models willaccount for energy policies and the desire for moreflexible SLAs, and will leverage new opportunities forintercloud federations.
We recommend research focus in designing anddeveloping the following important areas:
• Novel lightweight virtualization technologies forworkload specific accelerators, such as FPGAs andTPUs, that will proliferate the Cloud to Edge con-tinuum and facilitate low overhead serverless com-puting.
• System software of micro data centres for remotemanagement of the system stack and workload or-chestration.
• New power-aware strategies at the middleware andapplication levels for reducing energy consumptionof data centres.
• A unified marketplace for Fog/Edge computingthat will cater for competition between providersrather than consumers, and techniques for adopt-ing Fog/Edge computing rather than simply mak-ing devices and resource Fog/Edge enabled.
• Common standards for Cloud federation to allowusers to freely move between providers and avoidvendor lock-in. This includes, developing interoper-able systems that enforce user policies, manage thelifecycle of workloads, and negotiate Service LevelAgreements (SLAs).
• Mechanisms for vertically chaining network func-tions across different Cloud and Edge networks,multiple operators and heterogeneous hardware re-sources.
• Novel techniques for spatio-temporal compressionto reduce data sizes, adaptive indexing to managingindexes while processing queries, and developing‘NewSQL’ systems to support both transactionaland analytical workloads on the Cloud.
Although Cloud computing research has maturedover the last decade, there are numerous opportuni-ties to pursue meaningful and impactful research inthis area.
References
[1] B. Varghese and R. Buyya, “Next GenerationCloud Computing: New Trends and ResearchDirections,” Future Generation Computer Sys-tems, vol. 79, pp. 849 – 861, 2018.
[2] I. Foster and C. Kesselman, Eds., The Grid:Blueprint for a New Computing Infrastructure.San Francisco, CA, USA: Morgan KaufmannPublishers Inc., 1999.
[3] Y. Elkhatib, “Mapping Cross-Cloud Systems:Challenges and Opportunities,” in USENIXConference on Hot Topics in Cloud Computing.USENIX Association, Jun. 2016, pp. 77–83.
[4] D. Merkel, “Docker: Lightweight Linux contain-ers for consistent development and deployment,”Linux Journal, vol. 2014, no. 239, p. 2, 2014.
[5] I. Nadareishvili, R. Mitra, M. McLarty, andM. Amundsen, Microservice Architecture: align-ing principles, practices, and culture. O’ReillyMedia, Inc., 2016.
[6] C.-H. Hong, I. Spence, and D. S. Nikolopou-los, “GPU Virtualization and Scheduling Meth-ods: A Comprehensive Survey,” ACM Comput-ing Surveys, vol. 50, no. 3, p. 35, 2017.
[7] M. Satyanarayanan, “The Emergence of EdgeComputing,” Computer, vol. 50, no. 1, pp. 30–39, 2017.
9
[8] B. Varghese, N. Wang, S. Barbhuiya, P. Kil-patrick, and D. S. Nikolopoulos, “Challenges andOpportunities in Edge Computing,” in IEEE In-ternational Conference on Smart Cloud, 2016,pp. 20–26.
[9] S. J. Johnston, P. J. Basford, C. S. Perkins,H. Herry, F. P. Tso, D. Pezaros, R. D. Mullins,E. Yoneki, S. J. Cox, and J. Singer, “CommoditySingle Board Computer Clusters and Their Ap-plications,” Future Generation Computer Sys-tems, June 2018.
[10] Y. Elkhatib, B. F. Porter, H. B. Ribeiro, M. F.Zhani, J. Qadir, and E. Riviere, “On UsingMicro-Clouds to Deliver the Fog,” Internet Com-puting, vol. 21, no. 2, pp. 8–15, 2017.
[11] F. Yang and A. A. Chien, “ZCCloud: Explor-ing Wasted Green Power for High-PerformanceComputing,” in IEEE International Parallel andDistributed Processing Symposium, 2016, pp.1051–1060.
[12] J. Scheuner, P. Leitner, J. Cito, and H. Gall,“Cloud Work Bench - Infrastructure-as-CodeBased Cloud Benchmarking,” in IEEE Interna-tional Conference on Cloud Computing Technol-ogy and Science, 2014, pp. 246–253.
[13] B. Varghese, O. Akgun, I. Miguel, L. Thai,and A. Barker, “Cloud Benchmarking For Max-imising Performance of Scientific Applications,”IEEE Transactions on Cloud Computing, 2018.
[14] V. Dalibard, M. Schaarschmidt, and E. Yoneki,“BOAT: Building Auto-Tuners with StructuredBayesian Optimization,” in Proceedings of the26th International Conference on World WideWeb, 2017, pp. 479–488.
[15] L. Cui, F. P. Tso, and W. Jia, “Enforcing Net-work Policy in Heterogeneous Network Func-tion Box Environment,” Computer Networks,vol. 138, pp. 108 – 118, 2018.
[16] A. Elhabbash, G. Blair, G. Tyson, andY. Elkhatib, “Adaptive Service Deployment us-ing In-Network Mediation,” in InternationalConference on Network and Service Manage-ment, 2018.
[17] M. D. Assuno, R. N. Calheiros, S. Bianchi, M. A.Netto, and R. Buyya, “Big Data Computing andClouds: Trends and Future Directions,” Journalof Parallel and Distributed Computing, vol. 79-80, pp. 3 – 15, 2015.
[18] A. Pavlo and M. Aslett, “What’s Really Newwith NewSQL?” SIGMOD Record, vol. 45, no. 2,pp. 45–55, Sep. 2016.
[19] S. Hendrickson, S. Sturdevant, T. Harter,V. Venkataramani, A. C. Arpaci-Dusseau, andR. H. Arpaci-Dusseau, “Serverless Computationwith OpenLambda,” in 8th USENIX Workshopon Hot Topics in Cloud Computing, 2016.
[20] I. A. Kash and P. B. Key, “Pricing the Cloud,”IEEE Internet Computing, vol. 20, no. 1, pp. 36–43, 2016.
Author Biography
Blesson Varghese is a lecturer in computer science at Queens University Belfast and an honorary lec-turer at the University of St Andrews. His research interests are in tackling system oriented challenges ofnext-generation distributed systems that leverage the Cloud–Edge continuum. Varghese received a PhD incomputer science from the University of Reading. Contact him at [email protected].
Philipp Leitner is an assistant professor of software engineering at Chalmers and the University of Gothen-burg. His research interests are in software engineering for cloud- and web-based systems, with a specialfocus on software and service performance. Leitner received a PhD in business informatics from the ViennaUniversity of Technology. Contact him at [email protected].
10
Suprio Ray is an assistant professor at the University of New Brunswick. His research interests include bigdata, database management systems and data management in the Cloud. Ray received a PhD in computerscience from the university of Toronto. Contact him at [email protected].
Kyle Chard is a senior researcher and Fellow at the University of Chicago. His research interests includehigh performance computing, data-intensive computing, and cloud computing. Chard received a PhD incomputer science from Victoria University of Wellington. Contact him at [email protected].
Adam Barker is a Professor in Computer Science at the University of St Andrews. His research interestsare in scalable systems that address current and future large-scale data challenges. Barker received a PhDin Informatics from the University of Edinburgh. Contact him at [email protected].
Yehia Elkhatib is an assistant professor at Lancaster University and a visiting professor at L’Ecole deTechnologie Superieure, Montreal. He works to enable distributed applications to traverse infrastructuralboundaries, enforcing high-level application requirements, and involving associated interoperability and de-cision support issues. Elkhatib received a PhD in computer science from the University of Lancaster and isthe creator of the Cross-Cloud workshop series. Contact him at [email protected].
Herry Herry is a research associate at the University of Glasgow. His research interests include cloudcomputing, configuration management and AI. Herry received a PhD in cloud computing from the Universityof Edinburgh. Contact him at [email protected].
Cheol-Ho Hong is an assistant professor at Chung-Ang University. His research interests include operatingsystems and accelerator virtualization. Hong received a PhD in computer science from Korea University.Contact him at [email protected].
Jeremy Singer is a senior lecturer at the University of Glasgow. His research interests are in complexsystems engineering, focusing on the application of mathematical models to runtime system behaviour.Singer received a PhD in computer science from the University of Cambridge. Contact him at jeremy.
Fung Po Tso is a senior lecturer at Loughborough University, UK. His research interests include cloudnetworking, edge computing and network function chaining. Tso received a PhD in computer science fromthe City University of Hong Kong. Contact him at [email protected].
Eiko Yoneki is a research fellow at the University of Cambridge and a Turing fellow at the Alan TuringInstitute. Her research interests include distributed systems, large-scale graph processing, and computersystems optimisation with machine learning methods. Yoneki received a PhD in data centric asynchronouscommunication from the University of Cambridge. Contact her at [email protected].
Mohamed Faten Zhani is an associate professor at Ecole de Technologie Superieure, University of Quebec.His research interests include cloud computing, network and service management and large-scale distributedsystems. Zhani received a PhD in Computer Science from the University of Quebec. Contact him [email protected].
11