Performance analysis of parallel FDTD algorithm on different hardware platforms

Rapid developments in computer techniques and high performance networks have led to the widespread use of the parallel FDTD method for solving large EM problems. It is well known that the FDTD method is an embarrassingly parallel algorithm, which can make an efficient use of a computer cluster to reduce the computation time almost linearly as the number of nodes is increased, provided that the code is structured correctly and is run on the ‘right’ platform. In this communication, we investigate the parallel performance of the parallel FDTD method on different hardware platforms.