摘要
本文由基础的Linux操作系统入手 ,考察在一个SMP系统内部的两种不同的并行实现机制 :代表共享存储模型的线程模型 (和OpenMP模型 )和代表消息传递模型的MPI模型。然后 ,通过分析应当如何结合节点间和节点内两级并行得出 :从效率和易用性的综合考虑 ,在LinuxSMPCluster上应当直接使用利用共享内存进行通信的MPI进行编程。
This paper starts from an elementary operating system, describes two different parallel implementation mechanisms in an SMP system: the thread model ( and the OpenMP model) representing shared memory models,and the MPI model representing message passing models. It also analyses how to combine internode and intranode parallelisms, and concludes that on a Linux SMP cluster, MPI which communicates based on shared memory should be directly used to perform programming instead of joining other shared memory arrangements for the sake of efficiency and ease of use.
出处
《计算机工程与科学》
CSCD
2002年第4期78-80,88,共4页
Computer Engineering & Science
基金
国家自然科学基金资助项目 (6993 3 0 2 0
69873 0 0 4)
教育部跨世纪人才项目资助
