摘要
为了解决CFD技术在实际使用中碰到的计算资源紧张问题,提出了PC机局域网并行计算的思路.对LINUX操作系统、局域网的组建和网络并行环境MPI的实现进行了研究.搭建了并行计算平台,并利用商用CFD软件对该平台的基本工作性能进行了测试.结果表明:当网格数量超过50万时,两台计算机并行效率超过67.9%;四台计算机并行效率超过52.8%,随着网格数量增加,并行效率逐步得以提高,当网格数量接近200万时,并行效率分别达到168.58%和145.62%,出现了超线性现象.
In order to solve the problem of compute resource which arises in CFD technology, PC LAN with parallel computation was put forward. The LINUX operating system, construction of the local area network （I.AN） and implementation of the message passing interface （MPI） were studied in this paper. A parallel computation platform was built and its fundamental working performances were tested by commercial CFD software. The results show that the parallel efficiencies are above 67.9% for two-computer parallel system and 52.8% for four-computer parallel system when a computing grid larger than 0.5 million is used. With the increasing of the grid nodes, the parallel efficiency will be improved gradually. As the grid nodes are close to 200 million, the parallel efficiencies can reach 168. 58% and 145.62% respectively for the two parallel systems, and both present the superlinear behavior.
出处
《西安建筑科技大学学报（自然科学版）》
CSCD
北大核心
2007年第2期267-271,共5页
Journal of Xi'an University of Architecture & Technology(Natural Science Edition)
基金
国家自然科学基金项目(10272089)
关键词
CFD
并行计算
MPI
LINUX操作系统
parallel computation
message passing interface
LINUX operatingsystem
computational fluid dynamics
