7.1.2 Using Several Random-Number Generators



It is often advisable to design a simulation program so that it uses different independent random-number generators to produce different sequences of random numbers. This is easy to accomplish with most modern computer systems and enhances greatly the "controllability" of the experiments that one may wish to perform with a simulation program.

Consider, for example, the task of simulating demands for a spatially distributed urban service system. Assume that the times when such demands occur are statistically independent from the locations of these demands and that demands are served in a FCFS manner. Given then pdf's for the time distribution and for the spatial distribution of demands, it is advisable to use one random-number generator, A, to simulate the instants when demands occur and a different random-number generator, B, to simulate demand locations. For it may be desirable, in this case, to perform experiments in which, say, the times when demands occur change from one experiment to another while the sequence of the locations of demand remains the same. One way such experiments can be performed with two random-number generators is by using one single seed for B for all experiments while using a different seed for A for each experiment. Indeed, it is not unusual to find urban system simulations that use 10 or more independent random-number generators.