基于集群SPMD算法及演化计算并行研究  

Research for cluster-oriented spmd algorithm and parallelization of evolutionary computing

作　　者：罗俊[1] 雷咏梅[1] 

机构地区：[1]上海大学,上海200072

出　　处：《计算机工程与设计》2005年第10期2610-2613,共4页Computer Engineering and Design

基　　金：上海市教委发展基金项目(01A01)

摘　　要：高性能计算在科学研究领域有着广泛的应用。演化计算因具有计算规模大、种群中个体相关性小等优点,成为并行计算研究的主要对象之一。提出两种并行策略,对顺序GA(Genetic Algorithm)实现并行。首先使用主从模式对多种群协同遗传算法实现并行,在此基础上通过对算法进一步改进,实现了基于对等模式的并行演化计算,从而提高了算法可扩展性。比较了两种并行模式的各自特点,通过SPMD(Single Program Multiple Data)算法实现和基于上海大学“自强2000”高性能计算机上的实例验证,改进算法具有更好的可扩展性,更易于推广到网格环境。High performance computing is popular in science research field. With the character of large-scale and lose-correlation, the evolutionary computing has become one of the main research objects of parallel computing. Two parallelization techniques used for GA were addressed. First, the multi-population cooperating GA was parallelized by master-slave mode. Second, the former GA was improved and then parallelized by peer-to-peer mode, thus better extensibility was acquired. The differences between the two parallel modes were compared. Through paralleling the algorithm of SPMD （single program multiple data） type, it is proved that the improved algorithm provides better extensibility and is more suitable for grid environment.

关 键 词：并行计算 演化计算 SPMD MPI并行编程 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]