p PM 2 p Analyze Measure PI 2 PM 2 Analyze Validate

Performance Extrapolation is the process of evaluating the performance of a parallel program in a target execution environment using performance information obtained for the same program in a different execution environment. Performance extrapolation techniques are suited for rapid performance tuning of parallel programs, particularly when the target environment is unavailable. This paper describes one such technique that was developed for data-parallel C++ programs written in the pC++ language. The technique uses high-level event tracing of a n-thread pC++ program run on a uniprocessor machine together with trace-driven simulation to predict the performance of the program run on an n-processor machine. Our results show that even with high-level events, performance extrapolation techniques are effective in isolating critical factors affecting a program’s performance and for evaluating the influence of architectural and system parameters.