Innovative approaches to parallelizing finite-difference time-domain computations

We investigate the Linux implementation of the finite-difference time- domain (FDTD) algorithm on a Beowulf cluster of parallel workstations. We discuss synchronization and domain decomposition alternatives. To obtain performance characteristics the algorithm is applied to a test scattering problem. The computation has been done using varying numbers of identical dual-processor AMD Athlon MP- 1800 workstations with the Message Passing Interface (MPI) used to handle the inter- processor communications and the parallel runtime.