数值计算与计算机应用 2008, 29(1 ) 65-72 DOI:     ISSN: 1000-3266 CN: 11-2124/TP

本期目录 | 下期目录 | 过刊浏览 | 高级检索                                                            [打印本页]   [关闭]
论文
扩展功能
本文信息
Supporting info
PDF(535KB)
[HTML全文](0KB)
参考文献[PDF]
参考文献
服务与反馈
把本文推荐给朋友
加入我的书架
加入引用管理器
引用本文
Email Alert
文章反馈
浏览反馈信息
本文关键词相关文章
本文作者相关文章
PubMed

NAPA软件的并行优化

金君,乔楠,梁德旺

南京航空航天大学内流研究中心;INTEL中国软件中心;南京航空航天大学内流研究中心 南京 210016;北京 100000;南京 210016

摘要

大规模数值计算受到通信模式、并行算法、I/O速度等的多方面因素的制约,并行程序的好坏直接影响并行机性能的发挥,本文分别对上述影响并行性能的重要因素进行了分析并对NAPA软件进行了优化,测试中发现本文采用的并行算法性能比优化前提高了41.1%,此外,本文采用支持多视口的MPI I/O接口性能有明显提高.最后,本文分析了并行NAPA软件的可扩展性,并采用高超声速平板流动进行了测试,在Grid 97*49*49算例中,64个进程的情况下得到了较高的加速比(53.7)和并行效率(84%),表明,优化后的软件具有较好的并行效率和可扩展性.

关键词

PARALLEL OPTIMIZATION FOR NAPA CODE

Jin Jun (Internal Flow Research Center, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China) Qiao Nan (INTEL China Software Center, Beijing 100000, China) Liang Dewang (Internal Flow Research Center, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China)

Abstract:

The performance of distributed parallel systems is influenced by many factors, for example, communication mode, parallel algorithm, I/O speed. These factors were discussed and optimized in this paper. Especially for the parallel algorithm, the performance got 41.1% improved after the optimization in this paper. After that, we optimized the I/O for NAPA code based on MPI I/O, and got better performance. At last, in order to analysis the scalability, a series of tests of the hypersonic flat plate flow workload were conducted to identify the efficiency and speedups of the parallel code and in the case of Grid 97*49*49, when 64 processors were conducted, we got good speedup(53.7) and high performance(84%), the results show that this parallel code has good scalability.

Keywords:
收稿日期  修回日期  网络版发布日期  
DOI:
基金项目:

通讯作者:
作者简介:

本刊中的类似文章

Copyright 2008 by 数值计算与计算机应用