Enhanced distributed computing message passing strategies for FDTD

The finite difference time domain method (FDTD) solves Maxwell's equations by employing numerically and storage intensive computation to map the electric and magnetic fields within a finite volume as an explicit function of time. Distributed computation, using heterogeneous networks of computers, is a cost-effective way of applying FDTD; however interprocessor communication is the rate-limiting step. In particular, in most laboratories there is a surplus of unused, desktop CPU cycles during the evening and night. We have investigated ways of utilizing this resource by examining various communication harnesses for distributed computing and have for the first time in a distributed FDTD EM application, employed data packing mechanisms to alleviate the bottleneck caused by network latency and limited bandwidth availability. (C) 1999 John Wiley & Sons, Ltd.