Evaluation of finite element system architectures

Abstract

The finite element method is a powerful engineering analysis tool. However, the scope and size of the problems solved are limited by the capabilities and costs of computers currently used. New computer architectures with parallel processing capabilities can exploit the parallelism in the finite element method. Because of the large number of alternative hardware and software configurations and the high costs incurred in the development of such systems, a methodology is needed to compare proposed finite element systems without implementing either hardware or software. This paper presents such a methodology. System architectures are evaluated by simulating the execution of the finite element software on the hardware. Hardware is represented as a set of processors and resources, while the software is modeled as a set of computational tasks organized into an acyclic directed graph. The simulation is performed by scheduling routines. The result of the scheduling routines is a schedule consisting of task-processor assignments and task starting and finishing times. Also, processor and resource utilization levels are generated. These results provide a means of comparing proposed finit element systems.