网格计算和云计算 360 度比较原文.docx
《网格计算和云计算 360 度比较原文.docx》由会员分享,可在线阅读,更多相关《网格计算和云计算 360 度比较原文.docx(29页珍藏版)》请在冰豆网上搜索。
网格计算和云计算360度比较原文
CloudComputingandGridComputing360-DegreeCompared
1,2,3IanFoster,4YongZhao,1IoanRaicu,5ShiyongLu
foster@mcs.anl.gov,yozha@,iraicu@cs.uchicago.edu,shiyong@wayne.edu
1DepartmentofComputerScience,UniversityofChicago,Chicago,IL,USA
2ComputationInstitute,UniversityofChicago,Chicago,IL,USA
3Math&ComputerScienceDivision,ArgonneNationalLaboratory,Argonne,IL,USA
4MicrosoftCorporation,Redmond,WA,USA
5DepartmentofComputerScience,WayneStateUniversity,Detroit,MI,USA
Abstract–CloudComputinghasbecomeanotherbuzzwordafterWeb
2.0.However,therearedozensofdifferentdefinitionsforCloudComputingandthereseemstobenoconsensusonwhataCloudis.Ontheotherhand,CloudComputingisnotacompletelynewconcept;ithasintricateconnectiontotherelativelynewbutthirteen-yearestablishedGridComputingparadigm,andotherrelevanttechnologiessuchasutilitycomputing,clustercomputing,anddistributedsystemsingeneral.ThispaperstrivestocompareandcontrastCloudComputingwithGridComputingfromvariousanglesandgiveinsightsintotheessentialcharacteristicsofboth.
1100-MileOverview
CloudComputingishintingatafutureinwhichwewon’tcomputeonlocalcomputers,butoncentralizedfacilitiesoperatedbythird-partycomputeandstorageutilities.Wesurewon’tmisstheshrink-wrappedsoftwaretounwrapandinstall.Needlesstosay,thisisnotanewidea.Infact,backin1961,computingpioneerJohnMcCarthypredictedthat“computationmaysomedaybeorganizedasapublicutility”—andwentontospeculatehowthismightoccur.
Inthemid1990s,thetermGridwascoinedtodescribetechnologiesthatwouldallowconsumerstoobtaincomputingpowerondemand.IanFosterandotherspositedthatbystandardizingtheprotocolsusedtorequestcomputingpower,wecouldspurthecreationofaComputingGrid,analogousinformandutilitytotheelectricpowergrid.Researcherssubsequentlydevelopedtheseideasinmanyexcitingways,producingforexamplelarge-scalefederatedsystems(TeraGrid,OpenScienceGrid,caBIG,EGEE,EarthSystemGrid)thatprovidenotjustcomputingpower,butalsodataandsoftware,ondemand.Standardsorganizations(e.g.,OGF,OASIS)definedrelevantstandards.Moreprosaically,thetermwasalsoco-optedbyindustryasamarketingtermforclusters.ButnoviablecommercialGridComputingprovidersemerged,atleastnotuntilrecently.
Sois“CloudComputing”justanewnameforGrid?
Ininformationtechnology,wheretechnologyscalesbyanorderofmagnitude,andintheprocessreinventsitself,everyfiveyears,thereisnostraightforwardanswertosuchquestions.
Yes:
thevisionisthesame—toreducethecostofcomputing,increasereliability,andincreaseflexibilitybytransformingcomputersfromsomethingthatwebuyandoperateourselvestosomethingthatisoperatedbyathirdparty.
Butno:
thingsaredifferentnowthantheywere10yearsago.Wehaveanewneedtoanalyzemassivedata,thusmotivatinggreatlyincreaseddemandforcomputing.Havingrealizedthebenefitsofmovingfrommainframestocommodityclusters,
wefindthatthoseclustersarequiteexpensivetooperate.Wehavelow-costvirtualization.And,aboveall,wehavemultiplebillionsofdollarsbeingspentbythelikesofAmazon,Google,andMicrosofttocreaterealcommerciallarge-scalesystemscontaininghundredsofthousandsofcomputers.Theprospectofneedingonlyacreditcardtogeton-demandaccessto100,000+computersintensofdatacentersdistributedthroughouttheworld—resourcesthatbeappliedtoproblemswithmassive,potentiallydistributeddata,isexciting!
Soweareoperatingatadifferentscale,andoperatingatthesenew,moremassivescalescandemandfundamentallydifferentapproachestotacklingproblems.Italsoenables—indeedisoftenonlyapplicableto—entirelynewproblems.
Nevertheless,yes:
theproblemsaremostlythesameinCloudsandGrids.Thereisacommonneedtobeabletomanagelargefacilities;todefinemethodsbywhichconsumersdiscover,request,anduseresourcesprovidedbythecentralfacilities;andtoimplementtheoftenhighlyparallelcomputationsthatexecuteonthoseresources.Detailsdiffer,butthetwocommunitiesarestrugglingwithmanyofthesameissues.
1.1DefiningCloudComputing
ThereislittleconsensusonhowtodefinetheCloud[49].WeaddyetanotherdefinitiontothealreadysaturatedlistofdefinitionsforCloudComputing:
Alarge-scaledistributedcomputingparadigmthatisdrivenbyeconomiesofscale,inwhichapoolofabstracted,virtualized,dynamically-scalable,managedcomputingpower,storage,platforms,andservicesaredeliveredondemandtoexternalcustomersovertheInternet.
Thereareafewkeypointsinthisdefinition.First,CloudComputingisaspecializeddistributedcomputingparadigm;itdiffersfromtraditionalonesinthat1)itismassivelyscalable,
2)canbeencapsulatedasanabstractentitythatdeliversdifferentlevelsofservicestocustomersoutsidetheCloud,3)it
isdrivenbyeconomiesofscale[44],and4)theservicescanbedynamicallyconfigured(viavirtualizationorotherapproaches)anddeliveredondemand.
Governments,researchinstitutes,andindustryleadersarerushingtoadoptCloudComputingtosolvetheirever-increasingcomputingandstorageproblemsarisingintheInternetAge.TherearethreemainfactorscontributingtothesurgeandinterestsinCloudComputing:
1)rapiddecreaseinhardwarecostandincreaseincomputingpowerandstoragecapacity,andtheadventofmulti-corearchitectureandmodernsupercomputersconsistingofhundredsofthousandsofcores;
2)theexponentiallygrowingdatasizeinscientificinstrumentation/simulationandInternetpublishingandarchiving;and3)thewide-spreadadoptionofServicesComputingandWeb2.0applications.
1.2Clouds,Grids,andDistributedSystems
ManydiscerningreaderswillimmediatelynoticethatourdefinitionofCloudComputingoverlapswithmanyexistingtechnologies,suchasGridComputing,UtilityComputing,ServicesComputing,anddistributedcomputingingeneral.WearguethatCloudComputingnotonlyoverlapswithGridComputing,itisindeedevolvedoutofGridComputingandreliesonGridComputingasitsbackboneandinfrastructuresupport.Theevolutionhasbeenaresultofashiftinfocusfromaninfrastructurethatdeliversstorageandcomputeresources(suchisthecaseinGrids)toonethatiseconomybasedaimingtodelivermoreabstractresourcesandservices(suchisthecaseinClouds).AsforUtilityComputing,itisnotanewparadigmofcomputinginfrastructure;rather,itisabusinessmodelinwhichcomputingresources,suchascomputationandstorage,arepackagedasmeteredservicessimilartoaphysicalpublicutility,suchaselectricityandpublicswitchedtelephonenetwork.Utilitycomputingistypicallyimplementedusingothercomputinginfrastructure
(e.g.Grids)withadditionalaccountingandmonitoringservices.ACloudinfrastructurecanbeutilizedinternallybyacompanyorexposedtothepublicasutilitycomputing.
SeeFigure1foranoverviewoftherelationshipbetweenCloudsandotherdomainsthatitoverlapswith.Web2.0coversalmostthewholespectrumofservice-orientedapplications,whereCloudComputingliesatthelarge-scaleside.SupercomputingandClusterComputinghavebeenmorefocusedontraditionalnon-serviceapplications.GridComputingoverlapswithallthesefieldswhereitisgenerallyconsideredoflesserscalethansupercomputersandClouds.
Figure1:
GridsandCloudsOverview
GridComputingaimsto“enableresourcesharingandcoordinatedproblemsolvingindynamic,multi-institutionalvirtualorganizations”[18][20].Therearealsoafewkeyfeaturestothisdefinition:
Firstofall,Gridsprovidea
distributedcomputingparadigmorinfrastructurethatspansacrossmultiplevirtualorganizations(VO)whereeachVOcanconsistofeitherphysicallydistributedinstitutionsorlogicallyrelatedprojects/groups.Thegoalofsuchaparadigmistoenablefederatedresourcesharingindynamic,distributedenvironments.Theapproachtakenbythedefactostandardimplementation–TheGlobusToolkit[16][23],istobuildauniformcomputingenvironmentfromdiverseresourcesbydefiningstandardnetworkprotocolsandprovidingmiddlewaretomediateaccesstoawiderangeofheterogeneousresources.Globusaddressesvariousissuessuchassecurity,resourcediscovery,resourceprovisioningandmanagement,jobscheduling,monitoring,anddatamanagement.
Halfadecadeago,IanFostergaveathreepointchecklist[19]tohelpdefinewhatis,andwhatisnotaGrid:
1)coordinatesresourcesthatarenotsubjecttocentralizedcontrol,2)usesstandard,open,general-purposeprotocolsandinterfaces,and3)deliversnon-trivialqualitiesofservice.Althoughpoint3holdstrueforCloudComputing,neitherpoint1norpoint2isclearthatitisthecasefortoday’sClouds.ThevisionforCloudsandGridsaresimilar,detailsandtechnologiesusedmaydiffer,butthetwocommunitiesarestrugglingwithmanyofthesameissues.ThispaperstrivestocompareandcontrastCloudComputingwithGridComputingfromvariousanglesandgiveinsightsintotheessentialcharacteristicsofboth,withthehopetopaintalesscloudypictureofwhatCloudsare,whatkindofapplicationscanCloudsexpecttosupport,andwhatchallengesCloudsarelikelytofaceinthecomingyearsastheygainmomentumandadoption.Wehopethiswillhelpbothcommunitiesgaindeeperunderstandingofthegoals,assumptions,status,anddirections,andprovideamoredetailedviewofboth