基于分布式平台的FDTD并行算法

基于分布式平台开展一种新的时域有限差分（FDTD）并行算法研究,该算法基于VC++、CUDA5.0平台开发,调用Intel MPI 4.1.0库进行测试,在上海交通大学高性能计算中心图形处理单元（GPU）集群、上海超级计算机中心的“魔方”商用超级计算机以及国家超级计算济南中心的“神威蓝光”国产超级计算机等平台开展软件调试。通过对纯CPU、GPU以及CPU和GPU的混合测试,线程调度水平、核心函数处理速度得到明显提升,同时减少了通信执行时间比例,提高了加速比和并行效率,最后以2×2微带阵列为验证模型进行拓扑优化测试,结果证明该算法准确、有效。

FDTD parallel algorithm based on distributed platform

A new finite difference time domain (FDTD) parallel algorithm is developed based on distributed platform, which is based on VC++, CUDA5.0 development platform, calling Intel MPI 4.1.0 library for testing, developing software debugging on the platforms of high performance computing center graphics processing units (GPU) cluster in Shanghai Jiao Tong University, "Rubik's Cube" commercial super computer at Shanghai Supercomputer Center, and "Divinity Blue" domestic super computer at the National Supercomputing Center in Jinan. By pure CPU, GPU, and CPU and GPU hybrid test, thread scheduling level and kernel function processing speed improve significantly, while the proportion of the execution time of communication reduces, and the acceleration ratio and operation efficiency improve. Finally, the topology optimization of the model is verified by 2×2 micro-strip arrays. The results show that the algorithm is accurate and effective.