A new finite element concurrent computer program architecture

Abstract

Abstract A new computer program architecture for the solution of finite elemet systems using concurrent processing is presented. The basic approach involves the automatic splitting of an arbitrary spatial domain. Processors are dynamically re‐assigned during the several phases of an analysis. Direct and iterative solution strategies are considered. Computational algorithms for finite element dynamic analysis of large‐scale structural problems that exploit concurrent features of MIMD computers are implemented in modules around the basic architecture. Also, problems with localized non‐linearities are treated. A first implementation on a 32‐processor hypercube achieves efficiency rates of up to 90 per cent.