Implementation of an Efficient and Powerful Parallel Pseudo-random Number Generator

A reliable method of generating a high-volume of pseudo-random numbers is an essential requirement for any sort of stochastic modelling or Monte Carlo simulation work. We describe in detail the parallel PVM implementation of a pseudo-random number generator that combines the LeapFrog and Shuffling algorithms to create a single generator that preserves and extends the well-known randomness properties of these generators. The property of mutual disjointness between parallel sequences of random numbers (as guaranteed by the LeapFrog algorithm) is shown to only hold for one cycle of the base-generator being used. This limitation can be overcome by adding a shuffling routine which breaks up sequential correlations within each sequence. Hence the combined Shuffling-LeapFrog approach can be safely used beyond the first cycle of the base-generator while preserving the guarantee of producing mutually disjoint sequence of numbers. A PVM Master-Slave implementation of the method is described. 1 I..