基于前期工作的EOSS 算法,给出了扩展条件下的OpenMP 静态调度能量优化算法——改进的能量最优OpenMP 静态调度算法(improved energy-optimal static scheduling,简称IEOSS).该算法在原有EOSS 算法的基础上,建模了数据cache 失效造成的访存延迟对并行循环性能及能量的影响,选择最优调度块大小S^*,同时结合动态电压/频率调节,获得最小能量消耗.选择NPB3.2-OMP 的5 个程序进行模拟,以480 个处理器、6

This paper presents expanded hypothesis based OpenMP static scheduling energy optimization algorithm—Improved Energy-Optimal OpenMP Static Scheduling, IEOSS. Based on EOSS algorithm, IEOSS algorithm exploits the impact of memory access latency on performance and energy of parallel loop. Due to cache miss, the optimal chunk S^* scales down processors’ voltage/frequency and obtains the minimal energy consumption. Five programs from NPB3.2-OMP for further evaluation. Take 480 processors, 64-byte cache line as an example: with 5% performance loss, IEOSS algorithm improves the energy savings of EP, IS, FT, CG, MG by 10.15%, 4.49%, 81.66%, 2.32%, 10.11% compared with energy consumption of OpenMP chunk scheduling. This shows that IEOSS can effectively improve energy optimization by combining DVS with choosing optimal chunk size.

国家自然科学基金(60921062, 60903044); 国家高技术研究发展计划(863)(2008AA01Z110); 核高基重大软件专项(2009ZX01036-001-003)