高校科研团队Microwulf并行计算系统的搭建
详细信息 本馆镜像全文    |  推荐本文 | | 获取馆网全文
摘要
由于成本限制,高校科研团队的计算机多为普通型个人电脑(PC),无法执行涉及多节点的并行计算任务。基于Beowulf架构与可于大众电脑市场采购到的普通硬件设备搭建了一套四节点8CPU核Microwulf并行运算系统(命名为LXR01)。该系统各节点均安装Windows7与Linux操作系统,其中服务器节点利用NFS建立共享文件系统,分别以gcc、OpenMPI为主要编译环境和并行计算环境。HPL测试表明,LXR01系统的最大运算能力为39.93Gflops,其计算效率和成本效率分别为81.6%和Y-583.8/Gflops。同时,SPECFEM3D代码的计算实例证明LXR01系统能有效地解决并行计算问题。
On account of cost restriction,the computers of university research team are almost personal computers,which can not perform parallel computational tasks related to multiple computer nodes.Based on architecture of Beowulf and common hardware components bought from mass PC market,a Microwulf parallelism system with four nodes and eight CPU cores,named LXR01,was built.Windows7 and Linux operating system were installed on each node of the LXR01 system.The master node of LXR01 built shared file system with NFS,and adopted gcc and OpenMPI as main compiler and parallelism environment,respectively.The LXR01 system was benchmarked using HPL,from which 39.93 Gflops was max performance and the computational efficiency and cost efficiency of LXR01 system was 81.6% and Y-583.8/Gflops respectively.Meanwhile,it was concluded from parallel computational example of SPECFEM3D codes on LXR01 system that the LXR01 system can solve parallel computational problems effectively.
引文
[1]ADAMS J C,BROM T H.Microwulf:a beowulf cluster forevery desk[C]//Proceedings of the 39th SIGCSE technicalsymposium on Computer science education.Portland,OR,USA:ACM,2008:121-125.
    [2]STERLING T,BECKER D J,SAVARESE D,et al.Beo-wulf:A Parallel Workstation For Scientific Computation[C]//In Proceedings of the 24th International Conference onParallel Processing.CRC Press,1995:11-14.
    [3]丁海平,刘启方,黄勇,等.三维地震动场数值模拟并行计算系统[J].地震工程与工程振动,2004,24(02):19-22.
    [4]刘宾,刘广钟.基于Linux集群的高性能计算环境[J].辽宁工程技术大学学报,2006,25(S2):254-256.
    [5]SONZOGNI V E,YOMMI A M,NIGRO N M,et al.A par-allel finite element program on a Beowulf cluster[J].Ad-vances in Engineering Software,2002,33(7-10):427-443.
    [6]DMITRUK P,WANG L-,MATTHAEUS W H,et al.Scalable parallel FFT for spectral simulations on a Beowulfcluster[J].Parallel Computing,2001,27(14):1921-1936.
    [7]李贵明,俞国扬,罗家融.基于Linux的Beowulf集群的实现[J].计算机工程,2003,29(11):49-51.
    [8]祝永志,赵岩,魏榕晖.基于MPICH的Beowulf集群系统构建与性能评测[J].计算机工程与应用,2006(14):132-133.
    [9]肖明旺,许坚,车永刚,等.一个实用高性能PC集群的Linpack测试与分析[J].计算机应用研究,2004(09):183-184.
    [10]王晓英,都志辉.基于HPL测试的集群系统性能分析与优化[J].计算机科学,2005,32(11):231-234.
    [11]KOMATITSCH D,RITSEMA J,TROMP J.The Spectral-Element Method,Beowulf Computing,and Global Seismol-ogy[J].Science,2002,298(5599):1737-1742.

版权所有:© 2023 中国地质图书馆 中国地质调查局地学文献中心