2.3 Frequently Asked Questions

1. What are pseudo-random numbers? Numbers generated through an algorithm that appear to be random, when in fact, they are created by a deterministic process.

2. Why do we want to control randomness within simulation models? By controlling randomness, we can better ascertain if changes in simulation responses are due to factors of interest or due to underlying statistical variation caused by sampling. Do you think that it is better to compare two systems using the same inputs or different inputs? Suppose we have a work process that we have redesigned. We have the old process and the new process. Would it be better to test the difference in the process by using two different workers or the same worker? Most people agree that using the same worker is better. This same logic applies to randomness. Since we can control which pseudo-random number we use, it is better to test the difference between two model alternatives by using the same pseudo-random numbers. We use seeds and streams to do this.

3. What are seeds and streams? A random number stream is a sub-sequence of pseudo-random numbers that start at particular place with a larger sequence of pseudo-random numbers. The starting point of a sequence of pseudo-random numbers is called the seed. A seed allows us to pick a particular stream. Having multiple streams is useful to assign different streams to different sources of randomness within a model. This facilitates the control of the use of pseudo-random numbers when performing experiments.

4. How come my simulation results are always the same? Random number generators in computer simulation languages come with a default set of streams that divide the “circle” up into independent sets of random numbers. The streams are only independent if you do not use up all the random numbers within the subsequence. These streams allow the randomness associated with a simulation to be controlled. During the simulation, you can associate a specific stream with specific random processes in the model. This has the advantage of allowing you to check if the random numbers are causing significant differences in the outputs. In addition, this allows the random numbers used across alternative simulations to be better synchronized. Now a common question in simulation can be answered. That is, “If the simulation is using random numbers, why to I get the same results each time I run my program?” The corollary to this question is, “If I want to get different random results each time I run my program, how do I do it?” The answer to the first question is that the underlying random number generator is starting with the same seed each time you run your program. Thus, your program will use the same pseudo random numbers today as it did yesterday and the day before, etc. The answer to the corollary question is that you must tell the random number generator to use a different seed (or alternatively a different stream) if you want different invocations of the program to produce different results. The latter is not necessarily a desirable goal. For example, when developing your simulation programs, it is desirable to have repeatable results so that you can know that your program is working correctly.

5. How come my simulation results are unexpectedly different? Sometimes by changing the order of method calls you change the sequence of random numbers that are assigned to various things that happen in the model (e.g. attribute, generated service times, paths taken, etc.). Please see the FAQ “How come my results are always the same?” Now, the result can sometimes be radically different if different random numbers are used for different purposes. By using streams, you reduce this possibility and increase the likelihood that two models that have different configurations will have differences due to the change and not due to the random numbers used.