期刊文献+

并行程序通讯性能优化软件设计与实现

Design and Implementation of a Performance-Optimizing Software for Parallel Program
在线阅读 下载PDF
导出
摘要 提出一种并行程序通讯性能优化方法,对任务分配和路由产生的网络负载进行统计,帮助用户发现网络通讯中的不平衡,并进行优化。该方法已在一个Transputer互连的并行机PAR95上实现,效果很好。 In order to take full advantage of parallel processing, we in China need to raise speed up ratio as much as possible. For this purpose we develop a performance - optimiaing software for our distributed memory machine - PAR95 and it does help us to make easy adjustments on a CFD (computational fluid dynamics) parallel program for raising the speed- up ratio to approximately 28 in the case of 40 nodes.Fis. 1 shows our PAR95 for a single user environment, which consists of 40 nodes in a two dimensional mesh. Each node contains two CPUs: i860 and T805. The i860 is solely responsible for executing the single user's program. The function of T805 is to gather statistics about network traffiC and send it to the host, to be processed by the performance optimizing software. The processed results are graphically displayed, say as shown in Fig. 3a.Then easy adjustments are made to get a satisfactory graphiCal display (Fig, 3b), which shows that optimized performance is obtained.
作者 谢澎 康继昌
机构地区 西北工业大学
出处 《西北工业大学学报》 EI CAS CSCD 北大核心 1996年第2期281-284,共4页 Journal of Northwestern Polytechnical University
基金 国防科委"八五"预研课题
关键词 并行计算机 并行程序 程序性能 软件工具 distributed memory machine, parallel processing, performance optimizing software
  • 相关文献

参考文献2

  • 1谢澎,1993年
  • 2尹彦芝,C语言高级实用教程,1993年

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部