We report the openmpparallelimplementation of a finitedifferencetimedomain (FDTD) method for computational electrodynamics. We have identified several time-intensive procedures in the program and parallelized the...
详细信息
We report the openmpparallelimplementation of a finitedifferencetimedomain (FDTD) method for computational electrodynamics. We have identified several time-intensive procedures in the program and parallelized the major loops within them after careful examination. Different loop scheduling schemes have been used and tested in order to reduce computation times. The final parallelized version speeds up the computation by nearly a factor of four between the single processor and eight processor test runs on an SGI Origin 2000 parallel system. The speedup plateaus after eight CPUs, but we expect better scalability will be achieved if larger problem sizes are used. Besides the advantage of reduced execution times, our parallel program can also solve FDTD problems of much larger sizes than the sequential code due to much larger memory space available to us on parallel systems.
暂无评论