分块求解三角形线性方程组的一种分布式并行算法  

An Improved Parallel Algorithm for Block Solving Triangular Linear Systems on Distributed-memory Multicomputers

作　　者：张学波[1] 高佳 高立梅 

机构地区：[1]装备指挥技术学院信息装备系,北京101416 [2]空军二十三厂

出　　处：《装备指挥技术学院学报》2010年第1期114-117,共4页Journal of the Academy of Equipment Command & Technology

摘　　要：在分布式存储环境下，提出了一种基于“分块”思想的求解三角形线性方程组的分布式并行算法。该并行算法不仅成功运用了数据压缩存储技术，将方程组系数矩阵的存储规模由n^2减少到n（n＋s）/2≈n^2/2，而且较好地实现了计算与通信的相互重叠，从而减少了处理机间的通信开销。数值试验表明：该并行算法具有较高的加速比和并行效率。Under distributed-memory multicomputers, a parallel algorithm for solving triangular linear systems based on ＂block＂ idea is present in this paper. Not only data compressing storage tech- nology is adopted successfully, which makes the memory size of coefficient matrix of the systems from n^2 to n（n＋s）/2≈n^2/2 , but also overlapping computing and communication is used preferably to reduce the communication spending in this algorithm. The numerical experiments show that the speedup and the parallel efficiency of the algorithm are high enough.

关 键 词：三角形线性方程组 分布式并行算法 数据压缩存储 计算与通信 重叠 负载平衡 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]