计算机与现代化 ›› 2011, Vol. 1 ›› Issue (6): 160-5.doi: 10.3969/j.issn.1006-2475.2011.06.045

• 应用与开发 • 上一篇    下一篇

高性能计算平台的IO性能测试与分析

李 亮1, 聂瑞华2   

  1. 1.华南师范大学计算机学院,广东 广州 510631; 2.华南师范大学网络中心,广东 广州 51063
  • 收稿日期:2011-03-10 修回日期:1900-01-01 出版日期:2011-06-29 发布日期:2011-06-29

Test and Analysis of IO Performance on High Performance Computing Platform

LI Liang1, NIE Rui-hua2   

  1. 1.School of Computer, South China Normal University, Guangzhou 510631, China;2.Network Center, South China Normal University, Guangzhou 510631, China
  • Received:2011-03-10 Revised:1900-01-01 Online:2011-06-29 Published:2011-06-29

摘要: 在高性能计算平台上测试NFS和Lustre文件系统在大规模并行计算环境下的IO速率,根据测试结果分析该实验平台的IO瓶颈并提出改进方案;之后测试本地Cache、并行应用进行IO时的TransferSize和并行程序存取文件的FileSize等因素对分布式文件系统性能的影响,并根据实验结果提出如何合理地部署并行软件以有效地利用本地Cache,以及提出在编写并行程序时设置合适的TransferSize和FileSize以提高IO性能的建议。

关键词: 高性能计算, 并行IO性能, 并行文件系统, IO性能测试, Lustre文件系统

Abstract: This paper tests the performance of NFS and Lustre file system in the large scale distributed computing environment on high performance computing platform. According to the results, the IO bottleneck is known in this high performance computing platform and an advice is given to improve the IO performance to solve the bottleneck problem. Then this paper tests how the local cache, file’s size and transferSize affect the parallel filesystems’ IO performance, and proposes how to deploy the parallel applications and write the parallel programs efficiently to improve the IO performance. 

Key words: high performance computing, parallel IO performance, parallel file system, IO performance testing, Lustre file system