A short note on results of testing two random number generators

Abstract

Park and Miller argue for a minimal standard generator using the Lehmer model. Wichmann and Hill take a different track, and propose a three-tiered multiplicative model. In practice, real differences exist in their behavior, and this paper presents the results of some simple statistical testing and runtime comparisons. The results raise questions about the statistical behavior of the the Park and Miller minimal generator when creating random streams within a narrow range. The Wichmann and Hill generator produces statistically better results under the circumstances we tested, but takes approximately twice as long to run.