Distributed computation of electromagnetic scattering problems using finite-difference time-domain decompositions

Finite-Difference Time-Domain (FDTD) is an important numerical method for solving electromagnetic scattering problems. Solutions for these problems are computationally intensive, In addition, large problems require large amounts of memory, making distributed memory desirable. This paper describes distributed FDTD methods and their implementations on a distributed workstation network using PVM. We examine the parallelism of the methods and evaluate computing performance by studying load balancing, communication pattern variations, and scalability of the programs on the distributed network systems. This work should be relevant to anyone solving partial differential equations using finite-difference techniques.<<ETX>>