A Research On The Network Parallelization Of FDTD With MPI

The Finite-Difference Time-Domain method is a kind of electromagnetism field numerical algorithm, solving Maxwell equations in time domain directly. With the development of computer, it has gained great improvement and been used in more fields. But in order to gain a correct solution FDTD must comply with the condition when the object dispersed into lattices, i.e. the maximum dimension of grid must be less than the one-tenth of wave length. So with the increase of the frequency and dimension of object, larger memory and longer time are needed. But a single PC can hardly satisfy the requirement.As FDTD has a feature that a field variable in a grid just correlate with the variable around it, it can be run on a parallel computer when transited into parallel algorithm. MPI is a convenient realization of message passing among processes, which can be easily introduced in FDTD program written in C/C++ language. It can run on most of computers. So our study is based on the fact that MPI runs on a small PC group. This paper first give out the basic theory of FDTD and MPI, then expatiate the parallel algorithm of FDTD, including the algorithm in 1-D, 2-D, 3-D free space, Mur absorbing boundary condition, G-PML absorbing boundary condition, S-T field connecting boundary, sources etc. The equations are deduced, too. Then the way to use the four kinds of MPI communicating modes is analyzed when programming parallel FDTD. Last through programming, and running on the PC group, a good result is obtained. It showed the correctness, feasibility, and high efficiency of the algorithm.