Latest revision as of 13:32, 17 August 2020

Abstract

We present aspects of computer implementation of finite element structural analysis capabilities into a series of concurrent machines. We approach the implementation issues from the viewpoint that the design of data structures for finite element computations will play a crucial role in the constructs of computational modules, data transfer mechanisms among processes, and, consequently, desirable non-numerical and numerical algorithms. We account for the essential architectural features of local memory machines, shared memory machines and connection machines that are pertinent to finite element computations. We outline a concurrent FEM system that we advocate, and report our activities on system design and algorithm development. We report also on our 'hands-on' experience in concurrent finite element computations using some of the existing multiprocessors such as Intel's hypercube iPSC, Encore's Multimax, Alliant/FX8, and the CRAY as a benchmark case.