Design criterion for the statistics exchange control algorithms used in the statistical synchronization method

Abstract

The Statistical Synchronization Method (SSM) (Gy. Pongor, 1992) is a promising alternative to the conventional synchronization methods for parallel discrete event simulation. The conditions of the statistics exchange of the Statistical Synchronization Method are studied. A statistics exchange control algorithm is proposed that is based on prediction and synchronization point deletion. The so-called penalty functions are introduced. They are used to give a mathematical criterion that can be a measure of the goodness of the different statistics exchange control algorithms. Both analytical treatment and simulation show that there is a trade-off between the accuracy of the results and the achievable speed-up. According to the simulation results, we can get near-optimal results for a relatively wide range of values of the parameters of the prediction; thus SSM-T is robust enough to tolerate some inaccuracy of the parameters of the prediction.