Evaluating high-performance computing based on relative productivity indicator
Jie Wang, Huiying Lv, Yu Zeng, Yun Lin · 2013
Effective high-performance computing evaluation can promote the development of high-performance cluster systems tremendously. In this paper, we propose a reasonable and easy mechanism named RPI (relative productivity indicator) to evaluate high-performance cluster systems. RPI considers many factors comprehensively, such as system purchasing cost, operation cost, performance of key application, difficulty of programming and the complexity of management. RPI avoids the problem of different dimension of various parameters caused by direct measurement effectively. We also use a real high-performance cluster Dawning 5000A to prove the effectiveness of the RPI.