REX Recursive, Delta-Based Data-Centric Computation

REX Recursive, Delta-Based Data-Centric Computation

ID:39716468

大小:1.04 MB

页数:12页

时间:2019-07-09

REX Recursive, Delta-Based Data-Centric Computation_第1页
REX Recursive, Delta-Based Data-Centric Computation_第2页
REX Recursive, Delta-Based Data-Centric Computation_第3页
REX Recursive, Delta-Based Data-Centric Computation_第4页
REX Recursive, Delta-Based Data-Centric Computation_第5页
资源描述:

《REX Recursive, Delta-Based Data-Centric Computation》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、REX:Recursive,Delta-BasedData-CentricComputation∗SvilenR.MihaylovZacharyG.IvesSudiptoGuhaUniversityofPennsylvaniaPhiladelphia,PA,USA{svilen,zives,sudipto}@cis.upenn.eduABSTRACTqueriesforrelationshipanalysis,entityresolution,clustering,andrecommendation.Thishasspu

2、rredthedevelopmentofanewgen-Intoday’sWebandsocialnetworkenvironments,queryworkloadserationofscalable“NoSQL”clusterdataprocessingplatformsthatincludeadhocandOLAPqueries,aswellasiterativealgorithmsanalyzedataoutsidetheDBMS.ExamplesincludeGoogle’sMap-thatanalyzedata

3、relationships(e.g.,linkanalysis,clustering,learn-Reduce[8]anditsopen-sourceHadoopalternative,Pregel[20],ing).ModernDBMSssupportadhocandOLAPqueries,butmostDryad[16]andPig[21].Suchplatformsareoftentargetedatlarge-arenotrobustenoughtoscaletolargeclusters.Conversely,

4、“cloud”scale“cloudcomputing”tasks,hencewerefertothemascloudplatformslikeMapReduceexecutechainsofbatchtasksacrossplatforms,eventhoughtheycanbetargetedattraditionalclusters.clustersinafaulttolerantway,buthavetoomuchoverheadtosup-Thesecloudplatformssharemanyimplemen

5、tationtechniquesportadhocqueries.withparallelDBMSs,yettheymakedifferenttrade-offsthatyieldMoreover,bothclassesofplatformincursignificantoverheadinseveralbenefits:scale-uptomany(possiblyheterogeneous)nodes,executingiterativedataanalysisalgorithms.Mostsuchiterativeea

6、sierintegrationwithuser-definedcodetosupportspecializedal-algorithmsrepeatedlyrefineportionsoftheiranswers,untilsomegorithms,andtransparenthandlingoffailures(whichfrequentlyconvergencecriterionisreached.However,generalcloudplatformsoccurinlargeclusters).Ontheotherh

7、and,mostplatformssacri-typicallymustreprocessalldataineachstep.DBMSsthatsupportficeoneormoreof:high-levelprogrammingabstractions,prede-recursiveSQLaremoreefficientinthattheypropagateonlythefinedprimitiveslikejoins,ordeclarativeoptimizationtechniques.changesineachste

8、p—buttheystillaccumulateeachiteration’sMoreover,theseplatformstendtobemostusefulforbatchjobs,state,evenifitisnolongeruseful.User-definedfunctionsarealsoandsomea

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。