首页 | 本学科首页   官方微博 | 高级检索  
     

访存密集型应用在SMP机群系统中的性能分析
引用本文:顾丽红,吴少刚.访存密集型应用在SMP机群系统中的性能分析[J].小型微型计算机系统,2006,27(7):1258-1261.
作者姓名:顾丽红  吴少刚
作者单位:1. 中国石油大学(华东)计算机与通信工程学院,山东,东营,257061
2. 中国石油大学(华东)计算机与通信工程学院,山东,东营,257061;中国科学院,计算技术研究所,系统结构室,北京,100080
基金项目:中国科学院资助项目;科技部科研项目;中国石油大学校科研和教改项目
摘    要:SMP机群系统因其良好的性价比、卓越的可扩展性与可用性,逐渐成为当前高性能计算机领域的主流结构.这种结点内共享存储、结点间消息传递的两级混合结构是目前并行计算研究的热点,在单个SMP结点中,总线和内存带宽是否满足CPU和I/O的需求对于访存密集型应用的性能影响很大。本文针对访存密集型应用的特点测试分析了在SMP机群中访存冲突对系统性能的影响,结果表明我们的SMP结点存在性能瓶颈,这种量化分析对于设计大规模的基于SMP的机群系统有很好的指导意义.

关 键 词:访存密集应用  SMP机群  软件DSM系统
文章编号:1000-1220(2006)07-1258-04
收稿时间:08 12 2005 12:00AM
修稿时间:2005-08-12

Performance Analysis of Memory Intensive Applications on SMP Cluster
GU Li-hong,WU Shao-gang.Performance Analysis of Memory Intensive Applications on SMP Cluster[J].Mini-micro Systems,2006,27(7):1258-1261.
Authors:GU Li-hong  WU Shao-gang
Affiliation:1.College of Computer and Communication Engineering, University of Petroleum (East China;2. Computer System Lab, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080, China
Abstract:SMP clusters are the mainstream architectures in high performance computing because of their good cost performance ratio, excellent scalability and availability. A hybrid parallel programming paradigm combined intra-node shared-memory programming mode and inter-node message-passing programming mode has been becoming hot in research, In a single SMP node, whether system bus and memory bandwidths meet the CPUs and IO needs heavily affects the performance of memory-intensive applications. This paper evaluates the effect of memory access contention to performance of memory-intensive applications on a SMP cluster. The result shows that performance bottleneck exists in these SMP nodes of the SMP cluster due to deficient system bus and memory bandwidth. This quantitative analysis is beneficial to guiding large scale SMP Cluster design.
Keywords:OpenMP  JIAJIA
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号