矩阵在并行计算中的通信优化研究  

Research on Optimization of Large-scale Sparse Matrix Communication in Parallel Application

作　　者：卢珺[1] 

机构地区：[1]武汉大学东湖分校电子信息工程学院,湖北武汉430212

出　　处：《软件导刊》2009年第10期10-12,共3页Software Guide

摘　　要：为了提高并行应用系统的效率,研究了针对大型稀疏矩阵的压缩通信问题。通过对矩阵压缩通信过程中矩阵稀疏度、网络带宽、处理器计算能力之间的关系进行定量分析,推导出稀疏度下界计算公式。通过对不同稀疏度情况下算法所取得的效率进行分析,总结出压缩通信中稀疏度与通信效率之间的函数关系。结果表明本算法在稀疏矩阵通信方面效率有明显的提高。In order to improve the efficiency of parallel applications to study for a large sparse matrix compression communication problems. Communication through the process of compression matrix sparse matrix, the network bandwidth, processor computing power rela- tionship between the quantitative analysis, derived formula for calculating the lower bound of sparse degrees. Different degrees of sparse case the efficiency achieved by algorithm analysis, summarized in compressed sparse communication efficiency and communication between the degree of function. The results show that the algorithm in the sparse matrix communications improved efficiency.

关 键 词：并行计算 通信优化 油藏数值模拟 稀疏度 

分 类 号：TP301.4[自动化与计算机技术—计算机系统结构]