曙光5000A天体大规模数值模拟软件性能测试  

Performance Evaluation of the Simulation Software on Dawning 5000A for Large Scale Celestial Bodies

作　　者：王婷[1] 孙相征[1,2,3] 张云泉[1,2] 杨超[1] 李力刚[4] 刘芳芳[1] 管文华[1] 唐雨新[1,2,3] 姚继峰[5] 

机构地区：[1]中国科学院软件研究所并行计算实验室,北京100190 [2]中国科学院软件研究所计算机科学国家重点实验室,北京100190 [3]中国科学院研究生院,北京100190 [4]中国科学院上海天文台,上海200030 [5]上海超级计算中心,上海201203

出　　处：《西安交通大学学报》2009年第10期71-75,共5页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金重点资助项目(60533020);国家重点基础研究发展规划资助项目(2005CB321702);国家自然科学基金资助项目(60303020;10801125);国家高技术研究发展计划资助项目(2006AA01A102;2006AA01A125)

摘　　要：在国产百万亿次超级计算机曙光5000A上进行了天体大规模数值模拟软件的性能和可扩展测试实验,详细介绍了软件中的测试程序以及测试环境和过程,并对测试结果进行了分析.对于80×80×50的网格规模,采用每节点4进程测试了16～128个处理器核,每节点8进程、16进程分别测试了16～512个处理器核,相对加速比最终分别达到5.33、10.48和12.57,并行效率分别达到66.66%、32.58%和32.29%.对于160×160×100的网格规模,测试了每节点16进程的64～8192个核的性能,最大相对加速比为12.46,并行效率为9.73%.测试结果表明,曙光5000A具有良好的性能,测试结果对软件下一步的优化研究具有重要的指导意义.The performance and scalability evaluation of the large scale simulation software on planetary fluid dynamics are investigated by the experiments on the domestic 100 TFlops super- computer, Dawning 5000A. The details of the software testing procedures, the environment and the process of the testing are introduced. For the mesh size of 80 × 80 ×50, the performance testing on 16 to 128 processor cores with mapping 4 processes on each node and the performance testing on 16 to 512 processor cores with mapping 16 or 8 processes on each node are performed, respectively. The relative speedups reach 5. 33, 10. 48 and 12. 57, respectively, and the corresponding parallel efficiencies of these 3 testing cases are 66. 66%, 32.58% and 32.29%, respectively. For the mesh size of 160 ×160 ×100, the performance testing on 64 to 8 192 processor cores with mapping 16 processes on each node gives the results that the relative speedup is 12.46, and its parallel efficiency is 9. 73%. The experimental results reveal the good performance of Dawning 5000A and show important clues of the software optimization.

关 键 词：曙光5000A 数值模拟 性能测试 可扩展 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]