Parallelization of Monte Carlo Algorithms in Semiconductor Device Physics on Hypercube Multiprocessors
We have developed efficient parallel solutions of Monte Carlo algorithms for analyzing numerical models for charge transport used in semiconductor device physics. The algorithms were implemented on a 64 node hypercube multiprocessor and time measurements were made as both the problem size and number of processors are varied. A 64 node processor ensemble is measured to be 35 to 52 times as fast as a single processor when the problem size for the ensemble is fixed, and 61 to 63 times as fast as a single processor when problem size per processor is fixed. The latter measure, denoted scaled speedup, is shown to be better suited for denoting the parallel performance of Monte Carlo algorithms than the traditional measure of parallel speedup. Finally, an analysis of the test results are presented.
KeywordsProblem Size Parallel Performance Monte Carlo Algorithm Single Processor Parallel Speedup
Unable to display preview. Download preview PDF.
- J.L. Gustafson, G.R. Montry and R.E. Benner, “Development of parallel methods for a 1024-processor hypercube”, SIAM Journal on Scientific and Statistical Computing, Vol. 9, No. 4, July 1988Google Scholar
- G. Amdahl, “Validity of the single-processor approach to achieving large-scale computer capabilities”, AFIPS Conf. Proc, 30(1967), pp 483–485Google Scholar