Statistical Analysis of Simulation Output from Parallel Computing

Abstract

This article addresses statistical output analysis of transient simulations in the parallel computing environment with fixed computing time. Using parallel computing, most commonly used unbiased estimators based on the output sequence compromise. To rectify this issue, this article proposes an estimation procedure in the Bayesian framework. The proposed procedure is particularly useful when the computing time depends on the output value in each simulation replication. The effectiveness of our method is demonstrated through studies on queuing simulation and control chart simulation.