适合在分布存储大规模并行处理系统上应用的块 ADI 算法  

Block ADI Algorithm on Massively Parallel Processing System with Distributed Memory

作　　者：张宝琳[1,2] 徐涛[1,2] 杨烨[1,2] 乔香珍[1,2] 

机构地区：[1]北京应用物理与计算数学研究所计算物理实验室 [2]中国科学院计算技术研究所

出　　处：《高技术通讯》1998年第2期13-16,共4页Chinese High Technology Letters

基　　金：国家自然科学基金;863计划;中物院基金

摘　　要：提出了适合在分布存储的大规模并行处理系统上应用的块ＡＤＩ算法，实现了计算和通讯的局部化，大大减少了通讯花费。在曙光１０００计算机上的数值试验表明，块ＡＤＩ算法比传统的ＡＤＩ方法具有更高的并行计算效率。The traditional ADI method requires the expensive cost of communications on the distributed memory multiprocessors. To develop the difference methods which may easily implement on the distributed memory computers with a minimal cost of communications, the Block ADI method is given in the paper, which brings the localization strategy of both computation and communication a reality. Numerical experiments on Dawning 1000 computer show that the Block ADI method can get better efficiency of parallel computing than the traditional ADI method.

关 键 词：大规模 并行计算系统 分布存储 块ADI方法 

分 类 号：TP338.601[自动化与计算机技术—计算机系统结构]