Selecting random number seeds in practice

Abstract

A pseudorandom number generator, such as a multiplicative con gruential generator, depends on its initial value or seed. The com puter may select a seed using its internal clock. Alternatively the simulation analyst may use Fishman's tables with seeds spaced 100 000 apart. Theoretically, the concepts of sampling with and without replacement are involved. Practical problems arise if con secutive simulation runs cannot be made in a single terminal session. The user may select the simplest option. Runs for dif ferent systems may use common seeds to decrease the variability; then the internal clock is an impractical source since its inter nal (binary) representation must be saved.