Distributed QMC algorithms: new strategies and performance evaluation

摘要

We compare alternative strategies of a distributed algorithm to integrate a multidimensional function using Quasi-Monte Carlo techniques. We give a parallel performance analysis and experimental results, and introduce a new algorithm strategy that leads to near ideal speedup results, using up to 30 processors of a Beowulf type cluster. The results are obtained for a computational finance problem in 360 dimensions and for a 5-dimensional t-distribution arising in multiple comparison problems.