詹剑锋:BigDataBench——Benchmarking Big Data Systems

詹剑锋与大家分享了面向系统和体系结构的大数据基准测试程序的经验。他认为，大数据依赖于内存访问，成本的问题很重要。关于TPC，他总结到，TPC现在一半是技术，一半是政治。詹剑锋作为项目发起人在6个月之前进行的Benchmarking，初衷是为了将各种各样的英文术语进行统一。What is BigDataBench?An open source project on big databenchmarkinghttp://prof.ict.ac.cn/bigDataBench/3/.real- world data sets and 19 workloadsExtended in near future4V characteristicsVolume, Variety, velocity, and veracity中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENCA/ Comparison of big data benchmarking EffortsBenchmarkReal-world data setsData scalability Workloads variety Software stacksObjects to TestStatusEfforts(Volume, VeraeIt\HiBenchUnstructured text data( 1) PartialOffline analyticsHadoop+ HiveopOpenSciRceRealtime analyticsBigbenchNoneNAOffline analyticsDBMS and MapLe- DBMSand preMapReduceamp BigoneNARealtime Analytics RealtimeRealtime analyticpenScurceDataanalytic frameworks frameworksYCSBoneNAOnline servicesNoSQL systemsNoSQL SystemsOpenScurceLink benchUnstructured graph data(I) PartiOnline servicesGraph databaseGraph databaseOpenScurceCouldsuiteUnstructured text data(1) PartialOnline servicesNoSQL systemspencureOffine analyticsHadoopand ArchitecturegpanhlahBigDataBench Unstructured text data ( 1)Online servicesNOSQL systemsSystems andOpencurseSemi-structured text data(1)Offline analyticsDatabase.architectureUnstructured graph data(2)Realtime Analytics Realtimeand Nosql systemsOffieStructured table data(1)AnalyticsDifferent analyticsSemi-structured table data (1)frameworksframeworks「平新花哥本研兄5/Possible usersSystemsOS for big dataFile systems for big dataArchitectureDatrocessormanagementMemoryNetworksBiqData BenchPerformanceDistributed systemsoptimizationSchedulingCo-designProgramming systems中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENCResearch publicationsCharacterizing data analysis workloads in datacenters. Zhen Jia, Lei wang, jianfeng ZhanLixing Zhang and Chunjie luo. //SWC 2013Best paper award6/Big Data Bench: a Big Data Benchmark Suitefrom Internet Services. Lei Wang, jianfengZhan et al HPCa 2014, Industry session中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENCOutlineBenchmarking Methodology and DecisionCase Study3How to UseFuture work中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENCBig Data Bench Methodology4V of big dataBigData Bench中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENC9/Methodology(Cont)Data TvpesDataRepresent● StructuredSourcesBig dataative data●semi- structured● Text dataetsSets● Unstructured● Graph dats● Table dataPreserving● ExtendedInvestigateTypicaldata generation toolApplicationpreserving dataBigDataBenchDomainscharacteristicsBasic importantOperations andAlgorithmspplicationExtendedBig dataDiverseypesWorkloadsWorklo● Offline analyticsads● Realtime analyticsOnline servicesRepresentSoftware StackExtended*国綱字计算技求研完NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENC10/Methodology( cont4V of Big dataBig Data benchSystem and architectureSimilaritycharacteristicsanalysIs中国斜字氏计算技求研完呼NSTITUTE OF COMPUTING TECHNOLOGY, CHINESE ACADEMY OF SCIENC