HiC2009-Using_Simulation_for_Large-scale_Distributed_System

HiC2009-Using_Simulation_for_Large-scale_Distributed_System

ID:39712491

大小:566.23 KB

页数:16页

时间:2019-07-09

HiC2009-Using_Simulation_for_Large-scale_Distributed_System_第1页
HiC2009-Using_Simulation_for_Large-scale_Distributed_System_第2页
HiC2009-Using_Simulation_for_Large-scale_Distributed_System_第3页
HiC2009-Using_Simulation_for_Large-scale_Distributed_System_第4页
HiC2009-Using_Simulation_for_Large-scale_Distributed_System_第5页
资源描述:

《HiC2009-Using_Simulation_for_Large-scale_Distributed_System》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、MumakUsingSimulationforLarge-scaleDistributedSystemVerificationandDebuggingHongTang2009.11-HadoopWorld,BeijingHadoopPerformanceEngineering@Yahoo!ProductionCloudInvestigation-optimizationCycleAnalyticsBenchmarkingImplementationEvaluationToolsMetricsCollectionBenchmarkDevelopm

2、entDevelopmentOutlineMotivationsOverviewandStatusArchitectureLessonsandExperiencesConclusionsandFutureWorkMotivationsLarge-scaledistributedsystemishardtoverifyanddebug-Cannotafforda2000-nodeclusterforeverydeveloper,totesteveryfeatureenhancementorbugfix-Timeconsumingtor

3、unbenchmarks-HardtoreproducecornercaseconditionsMotivations(cont.)JobTrackerisafertileareaforexperimentation-Schedulingpolicies–wehavefourschedulersalready-SynergybetweenMapReduceandHDFS(e.g.blockplacementpolicies)-Improvingspeculativeexecutionpolicies-WEWANTMOREPEOPLETOHEL

4、PUSINNOVATE!But,JobTrackeristoocomplextomodifycorrectly-Manyfactorstoconsider:fairness,capacity/SLAguarantees,datalocality,loadbalance,failurehandlingandrecovery,etc-Manycontrolknobsincurrentimplementationwithsubtleinteractions–cantrickevenseasonedHadoopdevelopers-Generalad

5、vicetopeoplenewtoHadoopMapReducecodebase:DONOTPLAYWITHFIRE!Mumak–TheVisionDiscrete-eventsimulation-Cansimulateaclusterwiththousandsofnodesinoneprocess-DoesnotperformactualIOorcomputation-Virtualclock“spins”fasterthanwallclockPluggingintherealJobTrackerandSchedulercode-Good

6、:Noneedtoseparatelymodeltheschedulingpolicies-Bad/Good:InheritbothfeaturesandbugsinJTandSchedulerSimulateallconditionsofaproductioncluster-WorkloadandclusterconfigurationdrivenbytracegeneratedbyRumen-Jobsubmission,inter-arrival,jobdependencies,high-ramjobs-Allkindsoffailure

7、sandfailurerecoverylogic-ResourcecontentionProjectStatusWork-in-progressFirst-cutversioncommittedtoHadoop0.21andtrunk-Basicframework-Nomodelingofresourcecontention•Taskexecutiontimetakenas-isfromtraces-Onlyindividualtaskfailures,nonodefailuresorfailurecorrelations-Nojobdep

8、endencies-NospeculativeexecutionTheTeam-YahooHadoopPerformanceEngineeringT

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。
相关文章
更多
相关标签