Fast and reliable performance evaluation in mobile wireless networks using dynamic simulations

This paper presents statistical methods to enhance dynamic simulations in mobile networks. The aim is to perform reliable (unbiased) and fast evaluation of QoS indicators such as blocking and dropping rates. To do so, we adopt a trajectory splitting (TS) method that splits the path of the simulated stochastic process into multiple copies for determined configurations, thus focusing on configurations for which communication problems are likely to occur. We show that this method significantly reduces the variance of the estimators of both blocking and dropping probabilities, and hence reduces simulation times for the same precision. In a real network dynamic simulation, we propose an algorithm that automatically checks that the network has reached its steady state distribution and eliminates the bias due to the initial stabilization samples. We then isolate stations of interest with their most interfering neighbors and show that applying TS to these regions of interest leads to a very significant reduction of the simulation times.