摘要
研究了一种运行于PVM并行计算平台的矩阵相乘的并行算法。在工作站数量不为某个数的平方数时,Cannon算法在PVM环境下不能充分地利用机群系统中的资源。根据PVM并行编程环境中任务间通信的特点,文中设计了一种基于PVM的矩阵相乘并行算法,该算法根据工作站数量来确定子任务的数量,并对矩阵A进行分块,每个子任务可以计算一个分块。实验表明,该算法提高了机群并行环境中资源的利用率,提高了程序的运行效率。
This paper discusses a parallel algorithm for matrix multiple which run on the PVM. When the quantity of workstation is not for some integer square numbers, the Cannon algorithm can not fully use the resources. According to the features of task communication on the PVM, an effective parallel algorithm of matrix multiple is designed. Quantity of workstation determines that of task, in order to account a part for each task. Matrix A is compartmentalized. Experimental results show the algorithm improves resource utilization in fleet parallel environment, and improves operational efficiency of procedures.
出处
《计算机工程》
CAS
CSCD
北大核心
2007年第24期72-73,77,共3页
Computer Engineering
关键词
矩阵相乘
并行算法
机群并行系统
matrix multiple
parallel algorithm
fleet parallel environment
