Spark has emerged as an easy to use, scalable, robust and fast system for analytics with a rapidly growing and vibrant community of users and contributors. It is multipurpose - with extensive and modular infrastructure for machine learning, graph processing, SQL, streaming, statistical processing, and more. Its rapid adoption therefore calls for a performance assessment suite that supports agile development, measurement, validation, optimization, configuration, and deployment decisions across a broad range of platform environments and test cases. Recognizing the need for such comprehensive and agile testing, this paper proposes going beyond existing performance tests for Spark and creating an expanded Spark performance testing suite. This proposal describes several desirable properties flowing from the larger scale, greater and evolving variety, and nuanced requirements of different applications of Spark. The paper identifies the major areas of performance characterization, and the key methodological aspects that should be factored into the design of the proposed suite. The objective is to capture insights from industry and academia on how to best characterize capabilities of Spark-based analytic platforms and provide cost-effective assessment of optimization opportunities in a timely manner.
展开▼
