csinparallel pacific northwest 2016 workshop · 2016. 6. 14. · csinparallel.org take-home...
TRANSCRIPT
csinparallel.org
CSinParallel Pacific Northwest 2016 Workshop
Introduction to CSinParallel
Dick Brown, St. Olaf College
csinparallel.org
Workshopsite
OnCSinParallel.org:h7p://serc.carleton.edu/csinparallel/workshops/PacificNW16/index.htmlSeealsoworkshophandouts
csinparallel.org
Take-homemessages• Thetradi&onalCScurriculumis“strikingout”againstthefastballsandcurveballsofcurrentcompuGngtechnologies
• Ourstudentsneedprepara&oninparallelanddistributedcompu&ng(PDC)fortheircareers
• TheCSinParallelstrategyforeffecGve,quickercurricularchange
• StrategicresourcesforteachingPDCTheBIGchallenge:Forminganeffec&ve
communitytosupportthiscurriculartransi&on
csinparallel.org
StrikingoutagainstnewPDCtechnologies
h7p://armchairgm.wikia.com/File:1206127655_Strikeout.gif
csinparallel.org
Strike1:Mul&coreprocessors• Thenormforabout10years• Why?“HiWngthewall”in:
a. (Electrical)powerb. ILP(can’thidemuchmore
parallelismwithinacore)c. Deepeningmemory
hierarchy
StrikingoutagainstnewPDCtechnologies
h7p://images.bit-tech.net/content_images/2012/11/amd-fx-8350-review/piledriver-3b.jpg
csinparallel.org
Strike2:Cloudcompu&ngrevolu&on
StrikingoutagainstnewPDCtechnologies
h7p://cloudcompuGngadvices.com/wp-content/uploads/2012/08/cloud_compuGng-Features.jpg
• DistributedcompuGngempowersexciGngnewwebservices
csinparallel.org
Strike3:Heterogeneouscomputa&on(It’snotjustforHPCanymore)• CommoditychipswithmulGplecoretypes
– Intelvectorcores;AMDGPU+CPUcores
• MulGcoreandheterogeneousmulGcoreasdistributed/clustercompuGngnodes
StrikingoutagainstnewPDCtechnologies
csinparallel.org
EducaGngstudentsfortheircareers• ParallelismandconcurrencytradiGonallytaughtinArchitectureandOperaGngSystems– S&llnecessary,justnolongersufficient
• Commoditycompu&ng,notjustHPCresearch• HardwareandsoKwareevolvingrapidly
– Includinglanguages,libraries,frameworks• Recentcurriculumrecommenda&ons
– ACM/IEEECS2013,forundergraduateCSmajors– TCPPrecommendaGonsforPDC
Rapidcurricularchange??
csinparallel.org
EducaGngstudentsfortheircareers• ParallelismandconcurrencytradiGonallytaughtinArchitectureandOperaGngSystems– S&llnecessary,justnolongersufficient
• Commoditycompu&ng,notjustHPCresearch• HardwareandsoKwareevolvingrapidly
– Includinglanguages,libraries,frameworks• Recentcurriculumrecommenda&ons
– ACM/IEEECS2013,forundergraduateCSmajors– TCPPrecommendaGonsforPDC
…Rapidcurricularchange??
csinparallel.org
TheCSinParallelstrategy1. Brief,flexiblePDCteachingmodules• AlmostanyCScourse,atalmostanylevel• 1-to3-daycourseunits,forfeasibleincrementalmodificaGonstoasyllabus
• Broadvarietyoftopics,technologies,languages,etc.
• Adaptable/editableforlocalmodificaGon• Emphasisonhands-onexerciseswithcurrenttechnologies
• LearningobjecGves,teachingGps,etc.
csinparallel.org
TheCSinParallelstrategy2.Pedagogicaleffec&veness• Smallinterven5ons,bigimpact
– OpportuniGesforbroadexposuretoPDC• EffecGvehands-onlearningwithPDCtools• Spiralapproach
– RecurringtopicsinmulGplecontextsleadstobe7erretenGonanddeeperunderstanding
• Earlyandohen– Message:PDCisnaturalandpervasiveinCS
csinparallel.org
TheCSinParallelstrategy3.CommunityoffolksseekingtoteachPDC• Communitysupport
– CommunicaGngwithotherstryingsamethings(modules,insGtuGontypes,courses,techs,…)
– Resourcesforrecordingothers’pastexperiences(e.g.,Piazza,teachingGps)
Crea&ngcommunity/humannetworkingisaprimarygoalforthisworkshopevent
csinparallel.org
Somestrategicresources• Modules
– BasicCSinParallelmodules• Examples:WMRforCS1;MulGcoreProgramming(intro);ConcurrentDataStructures(C++orJava);ParallelSorGng
– “Exemplar”modules• PresentasignificantdomainapplicaGon(e.g.,drugdesign,epidemiology,trafficflow)togetherwithasequenGalimplementaGon
• Choiceofparallel/distributedimplementaGons• PedagogicalpossibiliGes+studentmoGvaGon
– “Taste”modules• Briefintroductoryexperienceofanewtechnology,orparallellanguage,largereducaGonalwork,etc.
csinparallel.org
Somestrategicresources• RelaGnglocalcoursesandCSinParallelmaterialstolatestcurriculumrecommenda&ons(CS2013,TCPP)
• ParallelProgrammingPaWerns– Recurringdesignstrategiesforparallelprograms,disGlledfrompracGcesofexperiencedpros
– GuidestoproblemsolvingandparallelthinkingforundergraduatelearnersofPDC
csinparallel.org
Somestrategicresources
• PlaXormresources– Example:WebMapReduce(WMR),forbeginningoradvancedstudentstolearnaboutscalablecomputaGonsthatfuelcloud-poweredservices
– Example:CDERhardwareresources,freeeducaGonalaccesstoPDChardware
– Example:Inexpensivemicro-clusters• Studentshelpingprofs,otherstudents
– ScalablecollaboraGveresource– Studentcapacityforexploringnewtechnologies
csinparallel.org
ThisworkshopAdaptableprogram–letusknowyourinterestsBasicplanfocusesonhands-onexperience
Tuesday1. CSinParallel(✓);PDC;resources2. OpenMP(shared-mem.parallel)3. WebMapReduce(XXXXLdata)
Wednesday1. MPI(distributedcompuGng)2. Somecurrenttrends3. GPU(SIMDparallel)
Thursday 1. IntegraGngPDCinyourcurriculum2. Nextsteps?
csinparallel.org
RefinedBIGchallenge
HowcanweformasupporGvecommunitythatbenefitsprofessorsmorethanit“costs”
Whatdowemeanby“costs”?• Learningnewsystems(e.g.,MPI,piazza)• Compe&&onfor&meduringtheterm• Weightofiner&aagainstchange
csinparallel.org
Recap• HardwareandsohwareparallelanddistributedcompuGng(PDC)technologiesareevolvingrapidly,withnoendinsight– MulGcore;cloud/distributed;heterogenous
• StudentsneedtoknowaboutPDCastheyentertheworkforce.
• CSinParalleloffersmodules,resources,andstrategiesforteachingPDC,inacontextofsuppor5vecommunity
• Biggestunsolvedproblem:Crea5ngasupportcommunityworththeeffort
csinparallel.org
QuesGons?