• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Conference on Parallel and Distributed Processing Techniques and Applications(PDPTA'03) v.2; 20030623-20030626; Las Vegas,NV; US >Communication Characteristics of a PC Cluster
【24h】

Communication Characteristics of a PC Cluster

机译：PC集群的通信特性

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

PC-clusters have become the fastest growing choice for building parallel platforms. The success of clusters has been mainly facilitated by the rapid advancement of microprocessor technologies and high-speed network interconnects such as Myrinet and Giganet. The increased accessibility and low cost of these technologies are also important factors for the wide acceptance of clusters among users. In PC cluster systems, the interconnection network connects low-cost hosts together using irregular topologies and provides the wiring flexibility and performance scalability. Cluster systems allow users to decide the system structure based on their needs as well as their available budget. MPI (Message Passing Interface) is the most popular communication interface used in today PC clusters and other cluster-type parallel computers, distributed computers, computational grids, as well as high-performance parallel computers [1]. There are several different open-source MPI implementations freely available and most of the implementations are based on the popular MPICH [2-3]. In general, these implementations have similar basic functions and they all produce almost identical performance. In this paper, widely used models and metrics for inter-processor communication costs are reviewed. Both synchronous and asynchronous modes for message completion are discussed. The experiment data from an OSCAR PC-cluster with MPICH using both modes of completion is modeled with multi-level least squared fitting techniques. Both results from point-to-point and collective communications are presented. Finally, some conclusions about the communication cost models on simple OSCAR clusters are drawn based on the experimental results.
机译：PC集群已成为构建并行平台的最快增长选择。集群技术的成功主要得益于微处理器技术的快速发展以及Myrinet和Giganet等高速网络互连。这些技术的增加的可访问性和低成本也是使群集被用户广泛接受的重要因素。在PC群集系统中，互连网络使用不规则的拓扑将低成本主机连接在一起，并提供了布线灵活性和性能可伸缩性。集群系统使用户可以根据自己的需求和可用预算来决定系统结构。 MPI（消息传递接口）是当今PC群集和其他群集类型的并行计算机，分布式计算机，计算网格以及高性能并行计算机中使用的最流行的通信接口[1]。有几种不同的开源MPI实现可免费获得，并且大多数实现基于流行的MPICH [2-3]。通常，这些实现具有相似的基本功能，并且它们都产生几乎相同的性能。在本文中，对处理器间通信成本的广泛使用的模型和度量进行了回顾。讨论了消息完成的同步和异步模式。使用多级最小二乘拟合技术对来自带有MPICH且使用两种完成模式的OSCAR PC群集的实验数据进行建模。提出了点对点和集体交流的结果。最后，基于实验结果，得出了关于简单OSCAR集群上的通信成本模型的一些结论。

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