什么是云計算What-is-Cloud-Computing

CloudComputingLecture#1

WhatisCloudComputing?

(andanintrotoparallel/distributedprocessing)JimmyLinTheiSchoolUniversityofMarylandWednesday,September3,2008ThisworkislicensedunderaCreativeCommonsAttribution-Noncommercial-ShareAlike3.0UnitedStates

See/licenses/by-nc-sa/3.0/us/fordetailsSomematerialadaptedfromslidesbyChristopheBisciglia,AaronKimball,&SierraMichels-Slettvet,GoogleDistributedComputingSeminar,2007(licensedunderCreationCommonsAttribution3.0License)1Source:/WhatisCloudComputing?Web-scaleproblemsLargedatacentersDifferentmodelsofcomputingHighly-interactiveWebapplications1.Web-ScaleProblemsCharacteristics:Definitelydata-intensiveMayalsobeprocessingintensiveExamples:Crawling,indexing,searching,miningtheWeb“Post-genomics”lifesciencesresearchOtherscientificdata(physics,astronomers,etc.)SensornetworksWeb2.0applications…Howmuchdata?WaybackMachinehas2PB+20TB/month(2006)Googleprocesses20PBaday(2008)“allwordseverspokenbyhumanbeings”~5EBNOAAhas~1PBclimatedata(2007)CERN’sLHCwillgenerate15PBayear(2008)640K

oughttobeenoughforanybody.MaximilienBrice,?CERNMaximilienBrice,?CERNThere’snothinglikemoredata!s/inspiration/data/g;(BankoandBrill,ACL2001)(Brantsetal.,EMNLP2007)Whattodowithmoredata?AnsweringfactoidquestionsPatternmatchingontheWebWorksamazinglywellLearningrelationsStartwithseedinstancesSearchforpatternsontheWebUsingpatternstofindmoreinstancesWhoshotAbrahamLincoln?

X

shotAbrahamLincolnBirthday-of(Mozart,1756)Birthday-of(Einstein,1879)WolfgangAmadeusMozart(1756-1791)Einsteinwasbornin1879PERSON(DATE–PERSONwasborninDATE(Brilletal.,TREC2001;Lin,ACMTOIS2007)(AgichteinandGravano,DL2000;RavichandranandHovy,ACL2002;…)2.LargeDataCentersWeb-scaleproblems?Throwmoremachinesatit!Cleartrend:centralizationofcomputingresourcesinlargedatacentersNecessaryingredients:fiber,juice,andspaceWhatdoOregon,Iceland,andabandonedmineshaveincommon?ImportantIssues:RedundancyEfficiencyUtilizationManagementSource:Harper’s(Feb,2008)MaximilienBrice,?CERNKeyTechnology:VirtualizationHardwareOperatingSystemAppAppAppTraditionalStackHardwareOSAppAppAppHypervisorOSOSVirtualizedStack3.DifferentComputingModelsUtilitycomputingWhybuymachineswhenyoucanrentcycles?Examples:Amazon’sEC2,GoGrid,AppNexusPlatformasaService(PaaS)GivemeniceAPIandtakecareoftheimplementationExample:GoogleAppEngineSoftwareasaService(SaaS)Justrunitforme!Example:Gmail“Whydoityourselfifyoucanpaysomeonetodoitforyou?”4.WebApplicationsAmistakeontopofahackbuiltonsandheldtogetherbyducttape?Whatisthenatureofsoftwareapplications?FromthedesktoptothebrowserSaaS==Web-basedapplicationsExamples:GoogleMaps,FacebookHowdowedeliverhighly-interactiveWeb-basedapplications?AJAX(asynchronousJavaScriptandXML)Forbetter,orforworse…Whatisthecourseabout?MapReduce:the“back-end”ofcloudcomputingBatch-orientedprocessingoflargedatasetsAjax:the“front-end”ofcloudcomputingHighly-interactiveWeb-basedapplicationsComputing“intheclouds”Amazon’sEC2/S3asanexampleofutilitycomputingAmazonWebServicesElasticComputeCloud(EC2)RentcomputingresourcesbythehourBasicunitofaccounting=instance-hourAdditionalcostsforbandwidthSimpleStorageService(S3)PersistentstorageChargebytheGB/monthAdditionalcostsforbandwidthYou’llbeusingEC2/S3forcourseassignments!Thiscourseisnotforyou…Ifyou’renotgenuinelyinterestedinthetopicIfyou’renotreadytodoalotofprogrammingIfyou’renotopentothinkingaboutcomputinginnewwaysIfyoucan’tcopewithuncertainly,unpredictability,poordocumentation,andimmaturesoftwareIfyoucan’tputinthetimeOtherwise,thiswillbearichlyrewardingcourse!Source:/2007/05/page/2/CloudComputingZenDon’tgetfrustrated(takeadeepbreath)…ThisisbleedingedgetechnologyThoseW$*#T@F!momentsBepatient…ThisisthesecondfirsttimeI’vetaughtthiscourseBeflexible…TherewillbeunanticipatedissuesalongthewayBeconstructive…TellmehowIcanmakeeveryone’sexperiencebetterSource:WikipediaSource:WikipediaSource:WikipediaSource:WikipediaThingstogoover…CoursescheduleAssignmentsanddeliverablesAmazonEC2/S3Web-ScaleProblems?Don’tholdyourbreath:BiocomputingNanocomputingQuantumcomputing…Itallboilsdownto…Divide-and-conquerThrowingmorehardwareattheproblemSimpletounderstand…alifetimetomaster…DivideandConquer“Work”w1w2w3r1r2r3“Result”“worker”“worker”“worker”PartitionCombineDifferentWorkersDifferentthreadsinthesamecoreDifferentcoresinthesameCPUDifferentCPUsinamulti-processorsystemDifferentmachinesinadistributedsystemChoices,Choices,ChoicesCommodityvs.“exotic”hardwareNumberofmachinesvs.processorvs.coresBandwidthofmemoryvs.diskvs.networkDifferentprogrammingmodelsFlynn’sTaxonomyInstructionsSingle(SI)Multiple(MI)DataMultiple(MD)SISDSingle-threadedprocessMISDPipelinearchitectureSIMDVectorProcessingMIMDMulti-threadedProgrammingSingle(SD)SISDDDDDDDDProcessorInstructionsSIMDD0ProcessorInstructionsD0D0D0D0D0D1D2D3D4…DnD1D2D3D4…DnD1D2D3D4…DnD1D2D3D4…DnD1D2D3D4…DnD1D2D3D4…DnD1D2D3D4…DnD0MIMDDDDDDDDProcessorInstructionsDDDDDDDProcessorInstructionsMemoryTypology:SharedMemoryProcessorProcessorProcessorProcessorMemoryTypology:DistributedMemoryProcessorMemoryProcessorMemoryProcessorMemoryProcessorNetworkMemoryTypology:HybridMemoryProcessorNetworkProcessorMemoryProcessorProcessorMemoryProcessorProcessorMemoryProcessorProcessorParallelizationProblemsHowdoweassignworkunitstoworkers?Whatifwehavemoreworkunitsthanworkers?Whatifworkersneedtosharepartialresults?Howdoweaggregatepartialresults?Howdoweknowalltheworkershavefinished?Whatifworkersdie?Whatisthecommonthemeofalloftheseproblems?GeneralTheme?Parallelizationproblemsarisefrom:CommunicationbetweenworkersAccesstosharedresources(e.g.,data)Thus,weneedasynchronizationsystem!Thisistricky:FindingbugsishardSolvingbugsisevenharderManagingMultipleWorkersDifficultbecause(Often)don’tknowtheorderinwhichworkersrun(Often)don’tknowwheretheworkersarerunning(Often)don’tknowwhenworkersinterrupteachotherThus,weneed:Semaphores(lock,unlock)Conditio