PERFORMANCE OF ITERATIVE METHODS FOR DISTRIBUTED-MEMORY MACHINES

We consider iterative methods for the large linear systems arising fro m partial differential equation problems on nonrectangular domains. Th ese are somewhat banded but otherwise of no special structure beyond b eing sparse. We study the implementations and performance of several i terative methods on hypercube machines and we examine in detail the ef fects of communication and synchronization delays. Models are presente d for these computations and projections made about performance for ma ssively parallel machines. We conclude that it is currently difficult to achieve the potential of parallel machines for such methods. Some a pproaches for improving efficiency are suggested.