分布存储环境并行有限元研究进展

Progress of Parallel Finite Element Method on Distributed-memory Parallel Computer Systems

详细评述了近 1 0年来国内外分布存储环境下并行有限元的研究态势 .认为实现分布存储环境下有限元并行化的策略可以归结为基于并行数值分析方法和基于物理分割构造的并行有限元两类 .其中 ,前者的实现方式有直接利用并行数值分析工具软件包和分割系统方程并行求解的策略 ;后者的策略则有子结构法、单元逐集 (EBE)法以及单元重组 (FETI)法 .对于物理非线性和动力问题 ,现行的处理方法都是利用静力解法叠代计算实现 .以这些分析为基础 ,进一步分析了机群 (COW )环境下实现大规模并行有限元计算的可行性 .

The paper presents a detailed introduction on the tendency regarding parallel finite element couputing in distributed-memory system of the latest decade.It seems that current parallel methods of finite element can be achieved by means of parallel numerical method or constituting specific algorithm after physically dividing a given structure into typical sections in the light of domain-decomposition.In distributed-memory surroundings,for the former one,high quality software packages for parallel numerical analysis are available,and some simple parallel solvers for system equation through splitting it into several blocks have been developed as applicable algorithms.However,the most efficient and potential algorithms may be the latter,the representatives of which are parallel substructure approach,element-by-element approach and finite element tearing and interconnecting approach.It is obvious that those methods could be evolved for nonlinear and dynamic problems.The feasibility to realize large-scale parallel finite element in the circumstance of cluster-of-workstations (COW) is discussed in details on the evaluating index and capability.