| Finite Difference Time Domain (FDTD) method has such advantages as simple, easily to be grasped and used. So it becomes very popular and garners many attentions. But constrained by its stability conditions, large quantity of meshes are required for FDTD to simulate electrically-large-size or complicated-structure objects, and results in two big problems, e.g. consuming huge memory and long computer time. Those problems heavily obstruct the wide use of FDTD.With the rapid improvement of the personal computer and the local network, the parallel computing cluster, which uses a group of personal computers to perform parallel computation, is widely adopted. For a parallel computing cluster, computers are usually connected through a high-speed Ethernet and has higher cost-performance ratio. It becomes more and more popular in many fields including large-scale numerical computations.Parallel FDTD method combines the advantages of FDTD method and the parallel computing cluster, which provides a feasible method for the electrically-large-size or complicated electromagnetic calculations. Many literatures introduced various parallel FDTD methods, but up to now it still is a problem for FDTD to achieve high efficient parallel calculation. One commonly confronted problem is the parallel efficiency drops sharply with number of the computers. How to improve its parallel efficiency becomes one of the key factors for researches on parallel FDTD.This paper introduces a general three-dimension FDTD soft which employs uniform meshes , PML absorbing boundary condition and near-far field transformation. It's written in Visual Fortran and using MPI parallel library to simulate electrically-large-size or complicated-structure objects in parallel way. To achieve high effective parallel calculations, this paper studied and optimized some important steps that form the parallel FDTD, such as the area division to minimize the communication between computers, the network topology to realize the shortest corresponding distance, adopting self-define data type to eliminate data packing and unpacking performances, comparing the effectiveness of communication functions provided by MPI to search the optimum communication functions, inputting and outputting data in parallel, and etc. By above improvements, the efficiency of the parallel FDTD is elevated.To measure the efficiency of the proposed parallel FDTD software, two numerical experiments are conducted and tested on a parallel computing cluster composed of 16 PCs. The first is simulation the inner electromagnetic field of a rectangular waveguide using a parallel FDTD calculation (without PML). When 6 PCs were used in the parallel calculation, the speed-up and parallel efficiency are as high as 5.42 and 90% respectively. Even when up to 16 PCs were used, its speed-up and parallel efficiency could be maintained at about 11.69 and 73% respectively. The second experiment utilizes the general three-dimension parallel FDTD software to simulate an open-ended wave-guide. The results show that the speed up increased from 1.96 for 2 PCs and to 8 for 14 PCs, corresponding parallel efficiency of 14 PCs is 57.93%. All above measurement results demonstrate general three-dimension parallel FDTD software is able to carry out parallel FDTD calculations with high efficiency.
|
PDF Full Text Request