资源描述:
《grades-graphx_with_fonts.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、GraphX:AResilientDistributedGraphSystemonSparkReynoldS.Xin,JosephE.Gonzalez,MichaelJ.Franklin,IonStoicaAMPLab,EECS,UCBerkeley{rxin,jegonzal,franklin,istoica}@cs.berkeley.eduABSTRACTanddistributedsystems.Byabstractingawaythechallengesoflarge-scaledistributedsy
2、stemdesign,theseframeworkssimplifytheFromsocialnetworkstotargetedadvertising,biggraphscapturedesign,implementation,andapplicationofnewsophisticatedgraphthestructureindataandarecentraltorecentadvancesinmachinealgorithmstolarge-scalereal-worldgraphproblems.lear
3、ninganddatamining.Unfortunately,directlyapplyingexistingWhileexistinggraph-parallelframeworkssharemanycommondata-paralleltoolstographcomputationtaskscanbecumbersomeproperties,eachpresentsaslightlydifferentviewofgraphcompu-andinefficient.Theneedforintuitive,sca
4、labletoolsforgraphtationtailoredtoeithertheoriginatingdomainoraspecificfamilycomputationhasleadtothedevelopmentofnewgraph-parallelofgraphalgorithmsandapplications.Unfortunately,becauseeachsystems(e.g.,Pregel,PowerGraph)whicharedesignedtoefficientlyframeworkreli
5、esonaseparateruntime,itisdifficulttocomposeexecutegraphalgorithms.Unfortunately,thesenewgraph-paralleltheseabstractions.Furthermore,whiletheseframeworksaddresssystemsdonotaddressthechallengesofgraphconstructionandthechallengesofgraphcomputation,theydonotaddres
6、sthechal-transformationwhichareoftenjustasproblematicasthesubsequentlengesofdataETL(preprocessingandconstruction)ortheprocesscomputation.Furthermore,existinggraph-parallelsystemsprovideofinterpretingandapplyingtheresultsofcomputation.Finally,fewlimitedfault-t
7、oleranceandsupportforinteractivedatamining.frameworkshavebuilt-insupportforinteractivegraphcomputation.WeintroduceGraphX,whichcombinestheadvantagesofbothAlternativelydata-parallelsystemslikeMapReduceanddata-parallelandgraph-parallelsystemsbyefficientlyexpressi
8、ngSpark[12]aredesignedforscalabledataprocessingandarewellgraphcomputationwithintheSparkdata-parallelframework.Wesuitedtothetaskofgraphconstruction(ETL).Byexploitingleveragenewideasindistr