Optimizing the Parallel Implementation of a Finite Difference Time Domain Code on a Multi-user Network of Workstations

Abstract

The implementation of a parallel, three dimensional, finite difference time domain (FDTD) computer program is considered and applied to a test scattering problem on a multi-user network of desktop workstations. The computation has primarily been done on a local area network (LAN) using six identical HP 9000/715 workstations (i.e. a homogeneous environment) with the Parallel Virtual Machine (PVM) software being employed as the communications harness.In this paper the sequential and parallel FDTD approaches are reviewed. We investigate the factors which cause a reduction in efficiency in the latter, such as host allocation and load balancing. We propose a task migration process, which is efficient for the FDTD algorithm, as a partial solution. The advantages of this approach are discussed and further developments based on available computational resources are suggested