Generating parallel random sequences via parameterizing EICGs for heterogeneous computing environments

Abstract

Monte Carlo (MC) simulations are considered to be ideal for parallelization because a large Monte Carlo problem can often be easily broken up into many small, essentially independent, subproblems. Many Monte Carlo applications are suitable for grid computing environments. In such an environment, the number of substreams is not known in advance in the computing task. This is a challenge for generating random sequences by using the traditional splitting method, which is aimed at ways of partitioning the full period of a single sequence into parallel substreams. Explicit inversive congruential generator(EICG)[1] with prime modulus has some very compelling properties for parallel Monte Carlo simulations. EICG is an excellent pseudorandom number generator for parallalizing via parameterizing. This paper describes an implementation of parallel random number sequences by varying a set of different parameters instead of splitting a single random sequence. Comparisons with linear and nonlinear generators in the library: SPRNG[2] are presented. © 2010 Springer-Verlag Berlin Heidelberg.