CloudCmpComparingPublicCloudProviders

2024年3月17日發(作者：負摩阻力計算)

CloudCmp:ComparingPublicCloudProviders

,,a,

KatrinaLaCurts,MIT(6.897Spring2011)

Choosingapubliccloudproviderisadi?customermeanttocomparepublic

cloudproviderswheneacho?ersamyriadofrvices,instacetypes,andpricingschemes?Evenifa

customercouldmakeasystematiccomparison,wouldtheybeabletodiscernwhichcloudprovider

wasbest—intermsofcost-e?ectivenessorotherwi—foraparticularapplication?

CloudCmp,publishedatIMC2010,providesasystematicframeworkforcomparingcloudproviders.

heirmethodsandcompareAmazonAWS,RackspaceCloudServers,GoogleApp-

Engine,andMicrosoftAzure

1

.Ratherthancomparethedi?,whichpro-

gramminglanguagesareo?eredbyeachprovider,thetypeoftechsupportavailable),CloudCmp

measurestheperformanceofthecorervicesthatarecommontothefourproviders:theelas-

ticcomputecluster,persistentstorage(tablestorage,whicho?erskey/valuefunctionality;blob

storage,whichallowsurstostoreunstructureddata;andqueuestorage,whichisdesignedto

storemessagesthatarelaterpasdbetweencloudisntances),andnetworking(bothintra-cloud

andWAN).Theirmain?ndingisthatnoonecloudisbest;di?erentcloudsaresuitedfordi?erent

eci?cally,they?ndthefollowing:

?Badon?nishingtimesforwell-knownJavabenchmarks(SPECjvm2008),di?erentproviders

o?erdi?erentlevelsofcost-e?mple,alow-endinstanceofMicrosoftAzure

is30%moreexpensivethanthecomparableAmazonAWSinstance,butableto?nishtasks

er,higher-endinstanceswithinoneclouddonotnecessarilyo?er

ult,parallelapplicationsmightdowellto

umorelow-endinstancesratherthanafewhigh-endones.

?Theperformanceofdi?erentstoragetypes(table,blob,andqueue)

example,AWS’srespontimefortablequeriesisordersofmagnitudefasterthananyother

provider,butittakeslongerfortheirtablestoreachconsistency(around100msonaverage,

butsometimesashighas500ms;otherprovidernosuchinconsistencies).Blobstorage

andqueuestorageesimilarvariability.

?Thewide-areanetworklatencyroughlycorrespondstogeographiclocation,suggestingthat

themoregeographicallydiveracloudprovideris,hput,bothintra-

cloudandwide-area,isrelativelyconsistentacrossallproviders,exceptforRackspace,which

emstohavearatherunder-provisionednetworkinthemeasurements;itsintra-datacenter

TCPthroughputisbetween600and700Mbpslowerthanallotherproviders,anditsWAN

throughputisbetween100and200Mbpslowerthanallotherproviders.

Giventheresults,it’sstilldi?culttopicka“winning”iderisbest

overall,andit’r,theauthors

applytheresultsoftheirstudytospeci?mple,ane-commercewebsitedeals

withmanytableoperations,

ontheresultsfromCloudCmp,e-commercesitesshoulduAWS;ithasthebestperformanceon

tablestorage(thepaperveri?esthisconclusionbyrunningatofe-commerce-speci?cbenchmarks

oneachcloud).Thepapergivessimilarexamplesforparallelscienti?ccomputingapplicationsand

TheauthorslisttheirresultsintermsofcloudsC

1

,C

2

,C

3

,andC

4

.Wegiveourowninferencesastowhich

providereachC

i

correspondsto.

1

1

CloudCmp:ComparingPublicCloudProviders

,,a,

KatrinaLaCurts,MIT(6.897Spring2011)

i?ccomputingapplicationsareCPU-intensive,andthusshouldu

alow-endinstanceofMicrosoftAzure;Latency-nsitivewebsitesshoulduGoogleAppEngine,

whichhasthelowestwide-areanetworklatencydistribution.

Overall,CloudCmpisagood?turesit

measuresarelikelytobecommonamongstallcloudproviders,notjustthefourstudiedinthis

paper,andthecomparisontoreal-worldapplicationsishelpfulingivingmoreinsightintotheir

r,thepaperlacksthe“stresstest”,

MapReduce)’salsonotclearhowfarinthefuturethemeasurementswillhold,or

whetheraverageconsumerswould?ndthisperformancecomparisonasufulasasimplefeature

comparisonbetweencloudproviders.

2