Some Experimental Observations on the Behavior of Composite Random Number Generators

A series of experiments with composite random number generators utilizing shuffling tables is described. The factors investigated are: (1) the magnitude of the modulus (equivalently, the word-size of the machine), (2) the effect of the modulus value for the indexing generator, and (3) the table size used for shuffling. Experimental results indicate that: (1) on large word-length machines (permitting large modulus values), shuffling accomplishes little in comparison with selected simple generators, (2) on small word-length machines, shuffling can produce sequences having an increased period and demonstrating acceptable statistical behavior, and (3) a table size of 2 produces results comparable to those obtained with larger tables.