Parallel FDTD with openMP on large Shared Memory Processor
时间：03-26
整理：3721RD
点击：
Dear all,
Some time ago I converted the FDTD into a parallel form for trials on a 256 core SMP. The scalability of the parallel process is OK up to about 16 cores and then it rapidly flattens, as is shown in the attached graph. The reason for this flattening is probably a combination of excessive memory movement and SMP architecture related, yet it is very excessive.
I cannot find any similar results of the FDTD on a large number of SMP cores anywere. Has anyone else seen this sort of FDTD behaviour on large SMPs ? If so, could you add some reference please.
Some time ago I converted the FDTD into a parallel form for trials on a 256 core SMP. The scalability of the parallel process is OK up to about 16 cores and then it rapidly flattens, as is shown in the attached graph. The reason for this flattening is probably a combination of excessive memory movement and SMP architecture related, yet it is very excessive.
I cannot find any similar results of the FDTD on a large number of SMP cores anywere. Has anyone else seen this sort of FDTD behaviour on large SMPs ? If so, could you add some reference please.