Efficient simulation of high speed tandem networks using importance sampling and stochastic gradient techniques

Simulation using importance sampling (IS) can obtain large speed-up factors if the modification or bias of the underlying probability measures is properly chosen. In the paper, the stochastic gradient optimisation techniques previously used to analyse single queue networks with bursty traffic are extended to tandem networks of queues with bursty traffic. The effectiveness of IS and stochastic gradient algorithms is demonstrated by their application to the problem of estimating the cell blocking probability due to buffer overflow of each queue in a tandem configuration. For the examples presented, these methods achieve speed-up factors of 5 to 8 orders of magnitude over conventional Monte Carlo simulation.