Simulation to develop and test queue approximations: A case study

Abstract

We use simulation to both develop and test approximation for mulas for the operating charactertistics of a queue that occurs in queuing network models. In this context, the main simula tion issues are design of simulation experiments and evaluation of the difference between approximations and simulation esti mates. The simulation experiments described here have specific goals, use analytic results to reduce the number of factors, and are performed sequentially with initial experiments generating questions to address in later experiments. The difference between simulation estimates and approximations are evaluated by two summary measures: average absolute percentage error and a measure of simulation variability. In addition, another measure is examined: percent of cases where the approximation falls out side a 95% confidence interval computed from simulation esti mates. While this measure is commonly used by others, especially in graphs, work here suggests it is inappropriate.