伯克利云计算白皮书(英文全)

AbovetheClouds:ABerkeleyViewofCloudComputingMichaelArmbrustArmandoFoxReanGriffithAnthonyD.JosephRandyH.KatzAndrewKonwinskiGunhoLeeDavidA.PattersonArielRabkinIonStoicaMateiZahariaElectricalEngineeringandComputerSciencesUniversityofCaliforniaatBerkeleyTechnicalReportNo.UCB/EECS-2009-28(s).Allrightsreserved.Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforprofitorcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthefirstpage.Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspecificpermission.AcknowledgementTheRADLab'sexistenceisduetothegeneroussupportofthefoundingmembersGoogle,Microsoft,andSunMicrosystemsandoftheaffiliatemembersAmazonWebServices,CiscoSystems,Facebook,Hewlett-Packard,IBM,NEC,NetworkAppliance,Oracle,Siemens,andVMware;bymatchingfundsfromtheStateofCalifornia'sMICROprogram(grants06-152,07-010,06-148,07-012,06-146,07-009,06-147,07-013,06-149,06-150,and07-008)andtheUniversityofCaliforniaIndustry/UniversityCooperativeResearchProgram(UCDiscovery)grantCOM07-10240;andbytheNationalScienceFoundation(grant#CNS-0509559).AbovetheClouds:ABerkeleyViewofCloudComputingMichaelArmbrust,ArmandoFox,ReanGriffith,AnthonyD.Joseph,RandyKatz,AndyKonwinski,GunhoLee,DavidPatterson,ArielRabkin,IonStoica,andMateiZaharia([e-mail])UCBerkeleyReliableAdaptiveDistributedSystemsLaboratory:CloudComputing,UtilityComputing,InternetDatacenters,DistributedSystemEconomics1ExecutiveSummaryCloudComputing,thelong-helddreamofcomputingasautility,hasthepotentialtotransformalargepartoftheITindustry,makingsoftwareevenmoreattractiveasaserviceandshapingthewayIThardwareisdesignedandpurchased.DeveloperswithinnovativeideasfornewInternetservicesnolongerrequirethelargecapitaloutlaysinhardwaretodeploytheirserviceorthehumanexpensetooperateit.Theyneednotbeconcernedaboutover-provisioningforaservicewhosepopularitydoesnotmeettheirpredictions,thuswastingcostlyresources,orunder-provisioningforonethatbecomeswildlypopular,thusmissingpotentialcustomersandrevenue.Moreover,companieswithlargebatch-orientedtaskscangetresultsasquicklyastheirprogramscanscale,sinceusing1000serversforonehourcostsnomorethanusingoneserverfor1000hours.Thiselasticityofresources,withoutpayingapremiumforlargescale,isunprecedentedinthehistoryofIT.CloudComputingreferstoboththeapplicationsdeliveredasservicesovertheInternetandthehardwareandsystemssoftwareinthedatacentersthatprovidethoseservices.TheservicesthemselveshavelongbeenreferredtoasSoftwareasaService(SaaS).ThedatacenterhardwareandsoftwareiswhatwewillcallaCloud.WhenaCloudismadeavailableinapay-as-you-gomannertothegeneralpublic,wecallitaPublicCloud;theservicebeingsoldisUtilityComputing.WeusethetermPrivateCloudtorefertointernaldatacentersofabusinessorotherorganization,notmadeavailabletothegeneralpublic.Thus,CloudComputingisthesumofSaaSandUtilityComputing,butdoesnotincludePrivateClouds.PeoplecanbeusersorprovidersofSaaS,orusersorprovidersofUtilityComputing.WefocusonSaaSProviders(CloudUsers)andCloudProviders,whichhavereceivedlessattentionthanSaaSUsers.Fromahardwarepointofview,threeaspectsarenewinCloudComputing.1.Theillusionofinfinitecomputingresourcesavailableondemand,therebyeliminatingtheneedforCloudCom-putinguserstoplanfaraheadforprovisioning.2.Theeliminationofanup-frontcommitmentbyCloudusers,therebyallowingcompaniestostartsmallandincreasehardwareresourcesonlywhenthereisanincreaseintheirneeds.3.Theabilitytopayforuseofcomputingresourcesonashort-termbasisasneeded(e.g.,processorsbythehourandstoragebytheday)andreleasethemasneeded,therebyrewardingconservationbylettingmachinesandstoragegowhentheyarenolongeruseful.Wearguethattheconstructionandoperationofextremelylarge-scale,commodity-computerdatacentersatlow-costlocationswasthekeynecessaryenablerofCloudComputing,fortheyuncoveredthefactorsof5to7decreaseincostofelectricity,networkbandwidth,operations,software,andhardwareavailableattheseverylargeeconomiesTheRADLab’sexistenceisduetothegeneroussupportofthefoundingmembersGoogle,Microsoft,andSunMicrosystemsandoftheaffiliatemembersAmazonWebServices,CiscoSystems,Facebook,Hewlett-Packard,IBM,NEC,NetworkAppliance,Oracle,Siemens,andVMware;bymatchingfundsfromtheStateofCalifornia’sMICROprogram(grants06-152,07-010,06-148,07-012,06-146,07-009,06-147,07-013,06-149,06-150,and07-008)andtheUniversityofCaliforniaIndustry/UniversityCooperativeResearchProgram(UCDiscovery)grantCOM07-10240;andbytheNationalScienceFoundation(grant#CNS-0509559).1ofscale.Thesefactors,combinedwithstatisticalmultiplexingtoincreaseutilizationcomparedaprivatecloud,meantthatcloudcomputingcouldofferservicesbelowthecostsofamedium-sizeddatacenterandyetstillmakeagoodprofit.Anyapplicationneedsamodelofcomputation,amodelofstorage,andamodelofcommunication.Thestatistical