Single run optimization of discrete event simulations-An empirical study using the M/M/1 queue

Simulation modeling has been widely used to analyze complex stochastic systems, such as, to compute some performance measures of a modern manufacturing facility. Often, we are interested in optimizing these performance measures of the system with respect to some controllable parameters. Traditional methods to find an optimum of a simulation model usually require making a number of simulation runs, which can be computationally intensive. This study proposes a stochastic optimization method to optimize a simulation model in a single simulation run, with the potential of large savings in computational effort. Two algorithms based on this method are developed and evaluated empirically using an M/M/1 queue problem. Experimental results show that the algorithms, especially one of them, are promising and that this approach merits further investigation.