基于Linpack的高性能计算机集群的并行性能测试  

Parallel Performance Evaluation for High Performance Computer Cluster Based on Linpack

作　　者：余江洪[1] 肖燕成[2] 朱宗柏[3] 蔡永华[3] 肖金生[3] 

机构地区：[1]中国船舶重工集团公司第七一二研究所,武汉430064 [2]华中科技大学计算机科学与工程学院,武汉430074 [3]武汉理工大学,武汉430070

出　　处：《船电技术》2009年第5期56-60,共5页Marine Electric & Electronic Engineering

摘　　要：基于计算机集群的高性能并行计算,已成为广大工程与科学计算的有效手段。本文以DELL HPCC高性能计算机集群为硬件平台,以Linpack为并行性能测试软件,测试了不同计算机节点数或处理器数、最高Gflops值的矩阵规模N、求解矩阵的分块大小NB、处理器网格中水平方向处理器个数P、处理器网格中垂直方向处理器个数Q的并行性能。通过测试和分析来研究高性能计算机集群的并行性能及其优化策略。High performance parallel computation based on computer cluster has become an efficient method for most engineering and scientific computation. By using L inpack as software platform and DELL HPCC as hardware platform, this paper tests different computational nodes and processors, matrix scale N of maximum value of Gflops, block size NB of matrix, horizontal numbers of processor P, and vertical numbers of processor Q. It investigates the parallel performance and optimization strategy of high performance computer cluster based on the test.

关 键 词：计算机集群 高性能计算 并行效率 计算流体动力学Linpack 

分 类 号：TP393.18[自动化与计算机技术—计算机应用技术]