Variance-based sampling for simulating cycle time—throughput curves using simulation-based estimates

Simulation at several different traffic intensities is required when generating a cycle time–throughput (CT–TH) curve. Previous work has shown that a variance–based allocation using an asymptotic variance approximation results in more precise confidence intervals than those achieved through naïve sampling in which an equal amount of effort is allocated to each traffic intensity being investigated. Many systems for which these CT–TH curves are desired are too complex for an asymptotic variance approximation to be easily determined. This paper presents a fixed–budget variance–based sampling allocation procedure for the simulation of CT–TH curves using variance estimates of the sample mean calculated from pilot simulation runs. The proposed allocation procedure significantly improved the range of precision over the naïve allocation.