高盛边缘计算专题报告

EQUITYRESEARCH|October14,2018|:0PMETHeatherBellini,[e-mail]&Co.LLCGoldmanSachsdoesandseekstodobusinesswithcompaniescoveredinitsresearchreports.Asaresult,investorsshouldbeawarethatthefirmmayhaveaconflictofinterestthatcouldaffecttheobjectivityofthisreport.Investorsshouldconsiderthisreportasonlyasinglefactorinmakingtheirinvestmentdecision.ForRegACcertificationandotherimportantdisclosures,seetheDisclosureAppendix,orgotofiliatesarenotregistered/qualifiedasresearchanalystswithFINRAintheU.S.TheGoldmanSachsGroup,[e-mail]&[e-mail]&[e-mail]&Co.LLCAsmoredevicesgeneratemoredatafrommorelocations,computingisfacingaspeed-versus-scalechallenge.Thepubliccloudremainsunrivaledinitscomputeandstorageresources,butgettingdatathereandbacktakestime,andultimatelyislimitedbythespeedoflightandthesizeofinternet“pipes.”Incaseswherebigdatawillbeusedtodrivereal-timedecision-making,weseeanopportunityfor“edgecomputing”tobecomeakeyenablerandextensionofthepubliccloudbyputtingcomputeandstorageresourcesclosertothedeviceorsourceofdatageneration.Edgecomputingcouldunlocka$40bnincrementalmarket($100bninthebullscenario),includingarangeofnewapplicationsthatcanbetterdirectoperations—from“whentobrake”foraself-drivingtruckto“whentochangecourse”foranoildrillworkingmilesunderground.WeseeprovidersofacohesivepubliccloudandedgesolutionasbestpositionedandhighlightMicrosoftinparticular,inadditiontoAmazon,Pivotal,andVMware.Howedgecomputingwillaugmentthecloud&unlockreal-time,bigdataapplicationsTheCutting‘Edge’ofComputingCloudPlatforms,[e-mail]3Shifttothecloudcontinuesinearnest11Butcomputingispoisedtoshiftbacktoadecentralizedparadigm13Whatisedgecomputing?17Edgecomputingdemanddrivers22Sizingthepotentialmarketopportunityforvirtualizationandserveroperatingsystems28Edgecomputingvs.cloudcomputingperformance33Killerappsforedgecomputing44Winners&losers:edgecomputingcouldsustainarenaissanceinon-premisesoftware49Valuation&risks67DisclosureAppendix6814October20182GoldmanSachsCl[e-mail]mportant?Whiletheoverarchingthemeinsoftwarewillcontinuetobethecentralizationofcompute(i.e.themovingofworkloadsfromon-premisestopubliccloud),webelievethatcomputingattheedgewillplayanincreasinglyimportantrole,augmentingthecapabilitiesofpubliccloudandbringingresourcesclosertothesourceofdatageneration.Inedgecomputing,dataisprocessed,analyzed,andacteduponat(orcloseto)thesourceofdatageneration,asopposedtorawdatabeingsentdirectlytoapublicorprivatecloudtobeactedupon.Toaccomplishthis,edgecomputingaddsthecorebuildingblocksofpubliccloud–includingcompute,networking,andstorage–closertotheoriginofthedata,allowinginsightstobegeneratedandexecutedinreal-time.Incontrastwithcentrally-locatedtraditionalandpurpose-builton-premisedatacentersorprivateclouds,edgeserverscanbeplacedfarfromcentralizedcomputingcores–in(oraround)factories,airplanes,cars,oilrigs,orinconjunctionwithcellphonetowers.Inanedge+cloudworld,processingisthereforedividedbetweentheedgeandthecloud,andfundamentally,ourviewisthatedgecomputingiscomplementaryto(andnotasubstitutefor)thepubliccloud–movingallcomputetotheedgewouldresultindistributedandunmanageableclustersofchaosandforgothescalebenefitsofpubliccloud.Althoughpubliccloudhaseffectivelylimitlessresources,edgecomputinghasseveraladvantagesthatcannotbeeffectivelymatchedbythepubliccloud.Forinstance,latency(distancetothepubliccloud)andbandwidth(sizeofthepipeconnectedtothepubliccloud)remainissuesinmanyinstances.Forusecaseswherereactiontimeiscriticaltothesuccessoftheoverallsystem,thelatencyinherentwitharoundtriptothecloudviaahub-and-spokemodelmaybenotbeacceptable.Latencycanbeinfluencedbyaplethoraofuncontrollablefactors,includingthenetworkconnectivityofthelocation,thenetworkprovider,othernetworktraffic,aswellasthespecificregion,availabilityzone,anddatacenterthattheuserconnectsto.Additionally,thespeedofcomputeanddataprocessinghasfaroutclassednetworkbandwidth.Trulybigdatausecaseswillalsocreatemassivedatageneration,ordersofmagnitudeabovewhatcouldbetransmittedbacktothepubliccloud;infact,thesebigdatausecaseswillgeneratesufficientdatathatsimplystoringit,evenwiththeresourcesofthepubliccloud(assumingthatthedatacanbetransmittedthere),willbechallenging;edgecomputingwillenablethedatatobeprocessedimmediately,andonlyrelevantdataneedstobesentbacktothepubliccloudtobestoredandfurtherreasonedupon.Dependenceonpubliccloudforalldataprocessingandanalyticsmaynotbesuitableformanyusecases,particularlythosethatfeatureloworintermittentnetworkconnectivity,andwebelievethateven5Gmaynotbeadequatebandwidthformanyusecases.Finally,processingthedataonthedeviceoratedge,versusuploadingrawdatatothepubliccloud,canyieldsuperiorresultsforsecurityandprivacy,asthereareinherent