Large scale ADI-FDTD parallel computations

We present performance benchmark results for a parallel alternating direction implicit finite difference time domain (ADI-FDTD) solver with convolution perfectly matched layer boundary conditions on a high performance computer cluster using a message passing interface library. Parallel implementations of ADI-FDTD are necessary to increase the maximum solvable problem size. ADI-FDTD solvers can reduce the overall run time of simulations requiring very fine meshing, compared to conventional FDTD, by overcoming the Courant-Friedrichs-Lewy constraint on the time step. We show that with careful minimization of the communication overhead, a scaling efficiency of 60–80% can be obtained for even the largest 3-D simulation that we characterize (8 billion mesh cells).