计算机工程 ›› 2010, Vol. 36 ›› Issue (12): 269-271.doi: 10.3969/j.issn.1000-3428.2010.12.093

• 开发研究与设计技术 • 上一篇    下一篇

基于Trace的并行性能模拟任务任务映射方法

方建滨,徐传福,车永刚,翁玉芬,王正华   

  1. (国防科技大学计算机学院并行与分布处理国家重点实验室,长沙 410073)
  • 出版日期:2010-06-20 发布日期:2010-06-20
  • 作者简介:方建滨(1984-),男,硕士研究生,主研方向:计算机系统性能评测与优化;徐传福,助理研究员、博士研究生;车永刚,副研究员、博士;翁玉芬,硕士研究生;王正华,教授、博士生 导师
  • 基金项目:

    国家“863”计划基金资助项目(2007AA01Z116);国家自然科学基金资助项目(60603055)

Task Mapping Method for Parallel Performance Simulation Based on Trace

FANG Jian-bin, XU Chuan-fu, CHE Yong-gang, WENG Yu-fen, WANG Zheng-hua   

  1. (National Key Lab for Parallel and Distributed Processing, School of Computer Science,National University of Defense Technology, Changsha 410073)
  • Online:2010-06-20 Published:2010-06-20

摘要:

针对Trace驱动的并行性能模拟问题,提出基于Trace信息指导的映射方法CO-LP3M。CO-LP3M利用从Trace中提取的目标应用程序的通信特征,以宿主机物理进程间通信次数最小化为目标,兼顾计算负载均衡,生成并行模拟任务到宿主机的映射。对Jacobi3D和HPL两个程序进行实验改为:对HPL程序进行实验(注:此处本来是两个程序的,后来为了缩减篇幅就删掉了其中的一个),结果表明CO-LP3M可有效提高并行模拟性能,相对于常见的映射方式,模拟性能最多提高14.7%。在此基础上给出CO-LP3M的扩展技术SCO-LP3M。

关键词: 并行模拟, 任务映射, 通信优化, 负载平衡

Abstract:

Aiming at minimizing communication between host physical processes and balancing computation loads, this paper proposes a mapping method named Communication-Optimal LP to PP Mapping(CO-LP3M) which can obtain the communication pattern of object application programs based on trace generated in parallel performance simulation. Experimental results on High Performance Linpack(HPL) show that the method can increase simulation performance by up to 14.7% compared with common mapping methods. Based on CO-LP3M, scalability technology named SCO-LP3M is given.

Key words: parallel simulation, task mapping, communication optimization, load balance

中图分类号: