• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2010, Vol. 32 ›› Issue (12): 102-106.doi: 10.3969/j.issn.1007130X.2010.

• 论文 • 上一篇    下一篇

气象应用基准测试程序MM5 Benchmark并行性能优化

朱小谦,张卫民,宋君强,胡庆丰   

  1. (国防科学技术大学计算机学院,湖南 长沙 410073)
  • 收稿日期:2009-01-14 修回日期:2009-04-16 出版日期:2010-12-25 发布日期:2010-12-25
  • 通讯作者: 朱小谦
  • 作者简介:朱小谦(1974),男,湖南望城人,博士,副研究员,研究方向为高性能并行计算和数值天气预报;张卫民,博士,研究员,研究方向为并行计算、数值天气预报和网格计算;宋君强,研究员,博士生导师,研究方向为高性能并行计算和数值天气预报;胡庆丰,教授,研究方向为计算机性能评价和并行算法。
  • 基金资助:

    国家863计划资助项目(2008AA01Z137)

Parallel Computing Performance Optimization of the Typical Meteorological Application: the MM5 Benchmark

ZHU Xiaoqian,ZHANG Weimin,SONG Junqiang,Hu Qingfeng   

  1. (School of Computer Science,National University of Defense Technology,Changsha 410073,China)
  • Received:2009-01-14 Revised:2009-04-16 Online:2010-12-25 Published:2010-12-25

摘要:

由美国国家大气研究中心(NCAR)和宾夕法尼亚大学(PSU)共同开发的MM5模式是目前气象领域广泛采用的中尺度模式之一。由于MM5模式的广泛应用和计算特点,MM5 Benchmark成为了高性能计算领域衡量计算机实用性能的典型应用之一。本文介绍了MM5模式的并行计算方法,针对目前流行的Intel Itanium2 CPU特点,在性能分析的基础上,从编译选项、代码优化、区域分解、处理机拓扑结构等方面对程序性能进行优化。实际测试结果表明,在某国产巨型机上优化后的持续计算性能最大提高20%左右。

关键词: MM5 benchmark, Itanium2, 并行, 性能分析, 优化

Abstract:

MM5 that is developed by NCAR and PSU is one of the most widely used mesoscale weather forecasting model,and the MM5 benchmark is becoming one of the representative applications to measure the performance of computers applied in the meteorology field because of its typical algorithm and popularity.This paper studies the parallel computing algorithm of MM5.According to the characteristics of the Intel Itanium2 CPU architecture,based on a computing performance analysis,the MM5 parallel computational efficiency is optimized from the aspect of compiler options,code tuning,partitioning methods and processor topologies.The optimization results indicate that  the MM5 implementation performance can be increased by 20% on a certain HPC platform based upon the Intel Itanium2 CPU.

Key words: MM5 benchmark;parallel computing;performance analysis;optimization