Sequential Batch Means Techniques for Mean Value Analysis in Distributed Simulation

In this paper we stress the necessity of design of eecient procedures for constructing conndence intervals of mean values in a simulation environment of multiple replications in parallel. Considering the popular approach based on analysis of batch means, we implemented and investigated sequential variant techniques that can give us an attractive speedup in simulation experiments of dynamic complex systems, such as communication networks, and at the same time to guarantee the accuracy of the nal results. Empirical results showing statistical properties of the mean value estimators in such distributed environment are presented.