Acceleration of Finite Element Analysis by Hierarchical Cluster of Workstations. High Speed Parallel Solving Method for Stiffness Equation.

Abstract

In general, most of execution time for large scale forging simulations by rigid plastic finite element method is spent to solve linear equations called stiffness equation. For speedup of large scale forging simulations, it is necessary to solve stiffness equation at high speed. In this paper, we present the result of accelerated computation for stiffness equation by the parallelized conjugated gradient method on the distributed multiprocessor system called cluster of workstations. As a result, the cluster of workstations, which are inter connected hierarchically by high speed network, spend less communication time than the ordinary non-hierarchical cluster of workstations. By using high speed network TPDDI (Twisted Pair Distributed Data Interface) and three-class model of hierarchical cluster of workstations which has two layers of slave CPUs, we achieved 50% reduction of communication time and about fivefold speedup of execution time of the parallelized conjugated gradient method by sixteen slave CPUs.