谷歌Google与云计算PPT资料.ppt
《谷歌Google与云计算PPT资料.ppt》由会员分享,可在线阅读,更多相关《谷歌Google与云计算PPT资料.ppt(32页珍藏版)》请在冰豆网上搜索。
AComputingCloudBreakthroughsforCloudComputingGoogleAppsforCloudComputingGoogleInfrastructureforCloudComputingTheInternetFromHardwaretoCommunityTheInternet:
FromHardwaretoCommunityMySpaceFacebook开心网校内网WhatDoTodaysUsersWant?
AccessibilityAccessfromanywhereandfrommultipledevicesShareabilityMakesharingaseasyascreatingandsavingFreedomUsersdontwanttheirdataheldhostageSimplicityEasy-to-learn,easy-to-useSecurityTrustthatdatawillnotbelostorseenbyunwantedparties6TheInnovationAComputingCloudCloudComputing7AttributesofCloudComputing8DatastoredonthecloudSoftware&
servicesonthecloud-AccessviawebbrowserBasedonstandardsandprotocols-Linux,AJAX,LAMP,etc.AccessiblefromanydeviceHardwareCentricSoftwareCentricServiceCentricPersonalPCClientServerCloudComputing9BreakthroughsforCloudComputingBreakthroughsforCloudComputing10User-Centric1Task-Centric2Powerful3Intelligent4Affordable5Programmable6UserCentricDatastoredinthe“Cloud”Datafollowsyou&
yourdevicesDataaccessibleanywhereDatacanbesharedwithothersmusicpreferencesmapsnewscontactsmessagesmailinglistsphotoe-mailscalendarphonenumbersinvestmentsExample:
GMailJustawebbrowserandyouraccountwithpassword!
Onceyoulogin,thedeviceis“yours”.Datastoredonremoteserversinthe“cloud”(withlargecapacity)Beijing,ontravelSanFrancisco,MondayHome,WednesdayUseGoogleDocstoSolveaTaskAccessyourdocsfromanywhereChatwithothersinrealtimeChangesinstantlyappeartoothercollaboratorsTask=“Teacherscreatingadepartmentalcurriculum”CommunicationTaskEmail,Chat,Contacts,ChatHistoryTask:
CollaborateonSpreadsheetCommunicateChatwithotherseditingthespreadsheetTask:
CollaborateonSpreadsheetCollaborateInviteotherstocollaborateonthespreadsheetTask:
CollaborateonSpreadsheetPublishInviteotherstoviewthespreadsheetYoucanalsoeasilyorganizeallyourcommontasksCloudComputingisPowerful:
ItcandowhatnoPCcandoIsGoogleSearchfasterthansearchinWindows/Outlook/Word?
AndGoogleSearchmustbemuchharder.Howmuchstoragedoesittaketostoreallofthewebpages?
100Bpages*10Kperpage=1000Tdisk!
CloudcomputinghasatitsdisposalEssentiallyinfiniteamountofdiskEssentiallyinfiniteamountofcomputation(Assumingtheycanbeparallelized)Example:
GoogleSearchWebPageSearchUniversalSearchW1stGeneration:
eraofsinglesearchnotdiverse2ndGeneration:
eraofverticalsearchtoocomplex3rdGeneration:
aneraofUniversalSearchABCDEFromverticalsearchtouniversalsearchABCDEIntegrationofuserexperienceUniversalSearchExampleUniversalSearchExampleCloudComputingInfrastructure25GFSArchitectureGoogle48%MSN19%Yahoo33%Filesbrokenintochunks(typically64MB)MastermanagesmetadataDatatransfershappendirectlybetweenclients/chunkserversClientClientClientReplicasMastersGFSMasterGFSMasterC0C1C2C5Chunkserver1C0C2C5ChunkserverNC1C3C5Chunkserver2ClientClientClientClientClientClientTypicalCluster26SchedulingmastersGFSchunkserverSchedulerslaveLinuxMachine1Userapp2Userapp1GFSmasterLockserviceGFSchunkserverSchedulerslaveLinuxMachineNUserapp3Userapp2Userapp1GFSchunkserverSchedulerslaveLinuxMachine2Userapp3MapReduce27Morespecifically28Programmerspecifiestwoprimarymethods:
map(k,v)*reduce(k,*)*Allvwithsamekarereducedtogether,inorder.Usuallyalsospecify:
partition(k,totalpartitions)-partitionforkoftenasimplehashofthekeyallowsreduceoperationsfordifferentktobeparallelized29BigTableDistributedmulti-levelmapWithaninterestingdatamodelFault-tolerant,persistentScalableThousandsofserversTerabytesofin-memorydataPetabyteofdisk-baseddataMillionsofreads/writespersecond,efficientscansSelf-managingServerscanbeadded/removeddynamicallyServersadjusttoloadimbalance30BigTable:
BasicDataModelDistributedmulti-dimensionalsparsemap(row,column,timestamp)cellcontentsGoodmatchformostofourapplications“”ROWSCOLUMNSTIMESTAMPS“contents”BigTable:
SystemArchitectureClusterSchedulingMasterhandlesfailover,monitoringGFSholdstabletdata,logsLockserviceholdsmetadata,handlesmaster-electionBigtabletabletserverservesdataBigtabletabletserverservesdataBigtabletabletserverservesdataBigtablemasterperformsmetadataops,loadbalancingBigtablecellBigtableclientBigtableclientlibraryOpen()ThanksQ&
A