Prototyping Parallel FDTD Programs by Macro Data Flow Graph Analysis

The paper presents a method for optimization of computations involved in solving finite difference time domain (FDTD) problems which enable electromagnetic wave propagation modeling in irregular areas. The method assumes program execution paradigm typical for macro data flow graph (MDFG) program representation. The proposed method enables MDFGs optimization based on three constituent steps. The first step consists in an initial MDFG design based on arbitrary introductory simulation area partitioning. The second step consists in merging initial macro data nodes to take into account computational system limitations and to obtain the macro node number equal to the number of physical processors in the system. The third step consists in computational cells redeployment to reduce inter-node data transmissions and balance computational loads of processors. Experimental results on parallel computational speedup are presented obtained with assumptions of different inter-processor communication systems