LOAD BALANCING USING HETEROGENEOUS PROCESSORS FOR CONTINUUM PROBLEMS ON A MESH

Continuum type problems are typically mapped onto a mesh with an equal distribution of computational load. However, such continuum problems typically cannot take advantage of wraparound communication links due to boundary conditions. Processors on corners and edges have fewer com munication links, giving more capacity for computation, assuming it is desirable to synchronize computation. This paper is concerned with lo ad balancing under these conditions and the introduction of heterogene ous processors for load balancing. The mapping of a continuum problem is typically considered a mapping of homogeneous tasks to homogeneous processors and communication links. This paper considers improvements in processing speed when homogeneous tasks characteristic of continuum problems are mapped to a topology heterogeneous in computation and co mmunication, allowing the mesh array to overcome differences in proces sor communication time to take full advantage of processing speed. (C) 1996 Academic Press, Inc.