Heuristic Rules for Improving the Quality of Simulation Based on the Sequential Non-overlapping Batch Mean

It is generally accepted that sequential procedures for stochastic discrete-event simulation of communication networks are the only practical approach for controlling statistical errors of the final simulation results. A commonly used stopping rule is based on the relative statistical error, since the magnitudes of the point estimates do not need to be known beforehand. The relative statistical error is defined as the ratio of the half-width of the confidence interval to the point estimate of the performance measure. The simulation is stopped when this ratio reaches a satisfactorily low value. One of the problems of this approach is that the inhe ently random nature of the output data can cause an accidental, temporary satisfaction of the stopping rule, resulting in acceptance of a result based on too-small samples of output data. We propose rules of thumb for protecting the quality of the final results from fully automated sequential procedures. The goal is to reduce the probability of finishing too early. The effectiveness of these rules of thumb is quantitatively assessed on the basis of the results of coverage analysis of a method of sequential output data analysis in steady-state simulation of communication networks: non-overlapping batch means.