基于PVM的网络并行子结构共轭梯度法  

NETWORK PARALLEL SUBSTRUCTURE CONJUGATE GRADIENT METHOD BASED ON PVM

作　　者：余天堂[1] 姜弘道[1] 

机构地区：[1]河海大学土木工程学院,南京210098

出　　处：《工程力学》2001年第5期29-35,共7页Engineering Mechanics

摘　　要：网络并行环境是近年来国际上并行环境的一个重要方向，ＰＶＭ是当前最流行的支持异构或同构型网络并行计算的软件平台之一。本文采用子结构共轭梯度法研究了基于ＰＶＭ的网络并行有限元，该方法将有限元网格划分为ｎ个子结构，再将ｎ个子结构的数据分送给网上ｎ台可用微机，ｎ台微机并行形成和组集ｎ个子结构的劲度矩阵和荷载列阵，然后采用预条件共轭梯度法并行求解结点位移，最后ｎ台微机并行对ｎ个子结构进行应变和应力分析。该方法不需形成结构的总体劲度矩阵和荷载列阵，可同时迭代求出所有结点位移，且比一般的迭代法收敛要快。算例表明此种并行子结构共轭梯度法在网络上能获得较高的并行加速比。In recent years, network parallel environment is an important direction of parallel environment, PVM is one of the best popular software platforms which support heterogeneous or isogeneous network parallel computing. In this paper, a network parallel finite element method based on PVM is studied with substructure conjugate gradient method. In the method, the finite element grid is divided into n substructures, the information of n substructures is distributed to n microcomputers. The generation and formation of stiffness matrixes and loading arrays of n substructures are done by n microcomputers simultaneously. Preconditioned conjugate gradient method is employed to compute the nodal displacements, then each microcomputer determines the strains and stresses for its associated substructure. The method does not require the formation of global stiffness and loading array, the nodal displacements can be simultaneously obtained. The speed of convergency is faster than general superposition method. Examples show that the proposed parallel substructure conjugate gradient method can achieve better parallel speedup ratio for network.

关 键 词：网络并行环境 PVM 子结构法 预条件共轭梯度法 并行加速比 有限元 

分 类 号：O342[理学—固体力学] O241.82[理学—力学]