Stochastic hybrid simulation with applications to queuing networks

This paper deals with an extension to the hybrid simulation paradigm, i.e. the combination of event-driven simulation and analytical modelling, applied to packet telecommunication networks. In order to speed up the simulation only a small part of all packets, the foreground traffic, is processed in an event-driven way. On each arrival of a foreground packet, the waiting time of the packet is sampled from the virtual waiting time distribution function of the combined foreground and background traffic. This distribution function is stochastically modelled by the exact large deviations asymptotic of the virtual waiting time in a many sources regime. This novel methodology is not only valid for wired point-to-point queueing networks having a fixed transmission capacity, but it can also be applied to queueing networks for which the transmission capacity varies with the traffic load of all the elements in the network. The results obtained by the stochastic hybrid simulator are compared to full-blown event-driven simulations. An important reduction in simulation run-time is gained without sacrificing accuracy.