Feasibility of fluid event-driven simulation for ATM networks

We describe an ATM network simulator that uses Markov-modulated fluid models for the sources as well as fluid leaky buckets and fluid bandwidth schedulers. The generalized processor sharing (GPS) and "idling" bandwidth schedulers are described. We argue for the use of the idling scheduler over GPS. Based on this fluid model, a simulator for ATM networks has been developed. The simulator employs the well-known discrete event-driven approach. Finally, simulation results are given that, in particular, compare the performance of fluid and "cell-level" simulators. The experimental results indicate that while the fluid simulator is much faster for ATM networks with certain characteristics, some key issues still need to be addressed to widen the applicability of this approach.