清华云计算课件--分布式计算

《清华云计算课件--分布式计算》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、Lecture1–IntroductiontoDistributedComputingMassDataProcessingTechnologyonLargeScaleClustersSummer,2007,TsinghuaUniversityAllcoursematerial(slides,labs,etc)islicensedundertheCreativeCommonsAttribution2.5License.ManythankstoAaronKimball&SierraMichels-Slettvetfortheiroriginalve

2、rsion1OutlineIntroductiontoDistributedSystemsParallelization&SynchronizationFundamentalsofNetworkingRemoteProcedureCalls(RPC)TransactionProcessingSystems2IntroductiontoDistributedSystems3ComputerSpeedup4Moore’sLaw:“Thedensityoftransistorsonachipdoublesevery18months,forthesam

3、ecost”(1965)Image:Tom’sHardwareWhyslowdownhere?Then,Howtoimprovetheperformance?ScopeofProblemsWhatcanyoudowith1computer?Whatcanyoudowith100computers?Whatcanyoudowithanentiredatacenter?5DistributedProblemsRenderingmultipleframesofhigh-qualityanimation6Image:DreamWorksAnimatio

4、nDistributedProblemsSimulatingseveralhundredorthousandcharacters7HappyFeet©KingdomFeatureProductions;LordoftheRings©NewLineCinemaDistributedProblemsIndexingtheweb(Google)SimulatinganInternet-sizednetworkfornetworkingexperiments(PlanetLab)Speedingupcontentdelivery(Akamai)8Wha

5、tisthekeyattributethatalltheseexampleshaveincommon?PlanetLab9PlanetLabisaglobalresearchnetworkthatsupportsthedevelopmentofnewnetworkservices.PlanetLabcurrentlyconsistsof809nodesat401sites.CDN-Akamai10Parallelvs.DistributedParallelcomputingcanmean:Vectorprocessingofdata(SIMD)

6、MultipleCPUsinasinglecomputer(MIMD)DistributedcomputingismultipleCPUsacrossmanycomputers(MIMD)11ABriefHistory…1975-85ParallelcomputingwasfavoredintheearlyyearsPrimarilyvector-basedatfirstGraduallymorethread-basedparallelismwasintroduced12Cray2supercomputer(Wikipedia)ABriefHi

7、story…1985-95“Massivelyparallelarchitectures”startrisinginprominenceMessagePassingInterface(MPI)andotherlibrariesdevelopedBandwidthwasabigproblem13ABriefHistory…1995-TodayCluster/gridarchitectureincreasinglydominantSpecialnodemachineseschewedinfavorofCOTStechnologiesWeb-wide

8、clustersoftwareCompanieslikeGoogletakethistotheextreme(10,000nodeclusters)1