ITR: Intelligent High-Performance Computing on Toys

点击翻译按钮获取中文摘要

Game consoles, with price points below $300, performance rivaling or exceeding that of PCs, and graphics capabilities recently found only on high-end visualization supercomputers, are the vanguard of yet another computing generation . computing on toys. Moreover, market forces and fierce vendor competition (e.g., among Microsoft, Nintendo, and Sony) continue to fuel technical innovation and performance improvements on these game platforms, creating research and development incentives and deployment opportunities in new domains. This proposal outlines a research plan to assess the utility and performance of game systems for both scientific computing and high-resolution visualization. This assessment of game systems will be based on development of a suite of adaptive performance analysis tools that support both offline and online performance optimization and their application to a suite of scientific and visualization codes. This effort leverages proposed Red Hat and NCSA software enhancements to PlayStation2 Linux software. In addition we currently are negotiating with Sony to acquire and deploy a large PlayStation2 cluster at NCSA for experimental assessment and scientific visualization. Our computing on toys software research plan focuses on three areas: (a) offline, multilevel performance instrumentation of applications and system software, (b) online, adaptive selection of multi-version code execution, and (c) experimental assessment using large-scale scientific applications and visualization software. New hardware performance measurements, instrumented scientific and graphics libraries, and performance derivatives, all integrated with our SvPablo performance analysis infrastructure, will provide the requisite data to move iterative performance tuning from an ad hoc style to one based on intelligent feedback and suggestions. The offline version of this SvPablo extension will accept performance metrics from hardware, software, and library instrumentation and generates suggestions for tuning locations and options using a fuzzy logic rule base that embodies performance tuning suggestions for the Sony PlayStation2.