Towards credible and efficient network simulation experiments

摘要

Despite of even higher performance computers, quantitative steady-state simulation of even a moderate complex system takes very long time. There is always a necessity of efficient procedures that can control the runlength of stochastic simulation experiments. This paper compares two sequential batching-based procedures for estimating steady-state mean under Multiple Replications in Parallel. A second motivation to this investigation is that fixed-sample size procedures can yield inappropriate precision of the results. By using well designed sequential analysis procedures for analysing data carefully, one can guarantee a better quality od the conclusions drawn from the analysis. A practical application illustrates the feasibility of these procedures.