• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>World Multiconference on Systemics, Cybernetics and Informatics(SCI 2002) v.11: Computer Science II; 20020714-20020718; Orlando,FL; US >An Empirical Study of Processes-to-Processors Mapping on Small-Scale SMP Clusters for Message-Passing Efficiency
【24h】

An Empirical Study of Processes-to-Processors Mapping on Small-Scale SMP Clusters for Message-Passing Efficiency

机译：小规模SMP集群上进程间映射以提高消息传递效率的实证研究

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Typically, a High Performance Computing (HPC) cluster loosely couples multiple Symmetric Multi-Processor (SMP) platforms into a single processing complex. Each SMP uses shared memory for its processors to communicate, whereas communication across SMPs goes through the intra-cluster interconnect. By analyzing the communication pattern of processes, it is possible to arrive at a mapping of processes to processors to ensure optimal communication paths for critical traffic. This critical traffic refers to the communication pattern of the program, which can be characterized by either frequency or size (or both) of the messages. To find an ideal mapping, it is imperative to understand the communication characteristics of the SMP memory system, intra-cluster interconnection, and the Message Passing Interface (MPI) program running on a cluster. For our study, we configured two 8-node dual-processor clusters, interconnected with Fast Ethernet, Gigabit Ethernet, and Myrinet respectively. One cluster is based on Pentium III CPUs with 133 MHz FSB and PC133 SDRAM servers; the other is a Pentium IV Xeon CPUs with 400 MHz FSB Rambus-RAM workstations cluster. We ran High Performance Linpack (HPL) benchmark on the two clusters to demonstrate how re-mapping of processes to processors (or changing the order of processors used for running the program) can affect the overall performance on the two clusters. The mappings are based on the HPL program analysis obtained from running a MPI profiling tool. Our results suggest that the performance of HPL can be improved dramatically depending on the process mapping and the problem size, as well as the performance characteristics of the systems' shard memory and the cluster interconnect. Conversely, an ad hoc mapping can adversely affect the cluster performance.
机译：通常，高性能计算（HPC）集群将多个对称多处理器（SMP）平台松散地耦合到单个处理复合体中。每个SMP使用共享内存供其处理器进行通信，而跨SMP的通信则通过集群内部互连进行。通过分析进程的通信模式，可以实现进程到处理器的映射，以确保关键流量的最佳通信路径。关键流量指的是程序的通信模式，可以用消息的频率或大小（或两者）来表征。为了找到理想的映射，必须了解SMP内存系统的通信特性，集群内部互连以及群集上运行的消息传递接口（MPI）程序。在我们的研究中，我们配置了两个8节点双处理器集群，分别与快速以太网，千兆以太网和Myrinet互连。一个群集基于具有133 MHz FSB的Pentium III CPU和PC133 SDRAM服务器。另一个是具有400 MHz FSB Rambus-RAM工作站集群的Pentium IV Xeon CPU。我们在两个集群上运行了高性能Linpack（HPL）基准测试，以演示将进程重新映射到处理器（或更改用于运行程序的处理器的顺序）如何影响两个集群的整体性能。映射基于从运行MPI分析工具获得的HPL程序分析。我们的结果表明，取决于过程映射和问题大小以及系统的分片内存和群集互连的性能特征，可以显着提高HPL的性能。相反，临时映射可能会对群集性能产生不利影响。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号