基于Docker的MPI和OpenMP混合编程

Hybrid Programming of MPI and OpenMP Based on Docker

针对当前搭建集群并行系统复杂且耗时等问题,提出基于Docker搭建并行系统。介绍轻量级虚拟化技术Docker的核心概念和基本架构,并基于Docker技术在Linux平台上搭建集群并行开发环境。简要阐述并行计算的思想,叙述MPI和OpenMP并行计算的基本概念和特点,针对矩阵并行乘法的算法建立MPI和OpenMP的混合编程模型,并给出混合编程模型与MPI并行编程模型以及OpenMP并行编程模型的性能对比,分析出现差异的原因。基于该混合编程模型比较Docker与传统物理机两者搭建的并行系统的并行效率。

Building cluster parallel system has been proved to be complex and time-consuming,this paper proposes a scheme of building a cluster parallel system based on Docker in response to this problem,introduces the concept and architecture of Docker that is a lightweight virtualization technology,and builds a cluster parallel development environment on Linux platform based on Docker technology. The idea of parallel computing is described in brief. The basic concepts and characteristics of parallel computing in MPI and OpenMP are listed. A hybrid programming model of MPI and OpenMP is built for the algorithm of matrix parallel multiplication. The performance of the hybrid programming model is compared with the MPI parallel programming model and the OpenMP parallel programming model,then a brief analysis of the reasons for the performance differences is given. Finally,this paper compares the parallel efficiency of parallel system built by Docker with that built by traditional computer based on the hybrid programming model.