BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking Framework[1] May 17, 2014 BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 1 / 10
Outline 5 NEGATIVE POINTS 1 OBJECTIVE 2 SUMMARY 3 OBSERVATIONS 4 POSITIVE POINTS 6 PROBLEM S IDENTIFIED 7 FEEDBACK 8 References BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 2 / 10
OBJECTIVE OBJECTIVE BigOP is a part of open source big data benchmarking project BigDataBench which features the abstraction of reprenstative Operation Sets, workload Patterns, and prescribed texts. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 3 / 10
SUMMARY SUMMARY Today we have landed into an era of big data and thus we are facing problem of how to choose the right system for processing big data. Benchmarking is an optimal way for evaluation and comparsion of systems.bigop is an end-to-end system benchmarking framework which facilitates automatic generation of tests with comprehensive workloads for big data systems.in BigOP, benchmarking test is specified by a prescription of one or more applications.a prescription incorporates a subset of operations and processing patterns,a data set,a workload generation method and metrices. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 4 / 10
OBSERVATIONS OBSERVATIONS An end-to-end model and adequate level of abstraction in BigOP provides space for various system implementations and optimizations. Due to large volume of data, big data system consists of nodes as well as datacenters,thus communication must be considered in benchmarks. Big data processing operations are classified as- element operation, single-set operation and double-set operation. The pattern abstraction are classified as single operation,multi-operation and iterative operation. A prescription test consists of a subset of operations and processing patterns, a data set, a workload generation method, and the measured metrics. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 5 / 10
POSITIVE POINTS POSITIVE POINTS BigBench focuses on big data analytics which adopts TPC-DS as the basis and adds new data types like semi-/un-structured data,as well as non-relational workloads. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 6 / 10
NEGATIVE POINTS NEGATIVE POINTS BigBench targets only a specific big data application and it doesn t cover the variety of big data processing workloads. The workload of the benchmark is too simple to meet the various needs of data processing. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 7 / 10
PROBLEM S IDENTIFIED PROBLEM S IDENTIFIED The present big data benchmarks covers only a part of BigOP s abstraction of processing operations and patterns. Present benchmarks are not as flexible as BigOP. Present benchmarks do not include the iterative pattern. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 8 / 10
FEEDBACK FEEDBACK Benchmarking tests can be conducted using BigOP s abstracted operations and patterns.system users can prescribe tests aiming at a specific application while developers can carry out general tests. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014 Framework[1] 9 / 10
References References I [1] Y. Zhu, J. Zhan, C. Weng, R. Nambiar, J. Zhang, X. Chen, and L. Wang, Bigop: Generating comprehensive big data workloads as a benchmarking framework, arxiv preprint arxiv:1401.6628, 2014. BigOP:Generating Comprehensive Big Data Workloads as a Benchmarking May 17, 2014Framework[1] 10 / 10