A parallel computational model for the finite element method on a memory-sharing multiprocessor computer

Abstract

This paper describes a new computing model for finite element analysis in a multiprocessor environment. The model is based upon the application of successive projections of the finite element space onto its supspaces. A problem is effectively decomposed into a collection of related subproblems which can be distributed asynchronously among any number of multiprocessors. Critical issues of parallel generation of coefficients, storage allocation, memory interference, interprocess communication and load balancing are dealt with and optimized in the context of shared memory multiprocessor computers. When implemented on test problems with 10 4 or more elements, the algorithms provide almost perfect speedup using from 2 to 24 processors at a time.