作者: S. Pllana , T. Fahringer
关键词:
摘要: High-performance computing is essential for solving large problems and reducing the time to solution a single problem. Current top high-performance systems contain 1000's of processors. Therefore, new tools are needed support program development that will exploit high degrees parallelism. The issue model-based performance evaluation real world programs on scale addressed in this paper. We present Performance Prophet, which modeling prediction tool parallel distributed programs. One main contributions paper our methodology evaluate model. In addition, we describe method automatic model generation. have implemented Prophet Java C++. illustrate approach by simulating real-world material science program.