PC机群上共享存储与消息传递的比较  

Shared-Memory Versus Message-Passing on PC Cluster

作　　者：章隆兵[1] 吴少刚[2] 蔡飞[1] 胡伟武[1] 

机构地区：[1]中国科学院计算技术研究所,北京100080 [2]石油大学(华东)计算机与通信工程学院,山东东营257061

出　　处：《软件学报》2004年第6期842-849,共8页Journal of Software

基　　金：国家自然科学基金;中国科学院全国首届优秀博士学位论文作者基金;中国科学院计算技术研究所领域前沿青年基金~~

摘　　要：共享存储和消息传递是目前两种主流的并行编程模型.一般认为,消息传递的可编程性不及共享存储友好.OpenMP是目前共享存储编程的实际工业标准.机群OpenMP系统在机群上提供了OpenMP编程环境,具有易编程和可扩展的特点,但是其性能如何一直是关注的热点.以机群OpenMP系统OpenMP/JIAJIA和典型的消息传递系统MPI(message passing interface)为例,在分析两种并行编程环境各自特点的基础上,在PC机群上采用7个应用比较了二者的性能.实验结果表明,对于所测试的7个应用而言,OpenMP版本在8个处理机运行时的平均性能为MPI版本的81%,但是采用OpenMP编程却比MPI简单很多.Two parallel programming models of shared-memory and message-passing are widely adopted. The programmability of message-passing is poor, while that of shared-memory is good. The OpenMP Application Programming Interface is an emerging standard for shared-memory. OpenMP on cluster supplies an OpenMP computing environment on cluster of workstations or PCs, which combines the friendly programmability of shared-memory with the fine scalability of cluster. Taking 7 well-known parallel applications on a cluster of PCs, this paper compares the performance of OpenMP/JIAJIA, an OpenMP system on cluster, with that of MPI, a typical message passing system. Experimental results show that the performance of OpenMP is averagely equal to 81% of MPI for the 7 applications running on 8-nodes, but the former is easier to use than the latter.

关 键 词：OPENMP 消息传递 PC机群 并行编程模型 共享存储 软件分布式共享存储系统 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]