The authors propose a novel hybrid technique for simulation of cell- or packet-switched networks using buffer/server models. Network traffic is divided into a foreground and a background. The analysis focuses on the buffer states, and is carried out by substituting buffer/server models that approximate the behavior of the buffers with the background traffic. Where the background traffic is much more voluminous than the foreground, cell-level simulation is increased by a similarly large factor in comparison with cell-level simulation of the traffic as a whole.
The authors obtain original results in queueing theory to derive the approximating buffer models, which are applied to two distinct regimes. The first is for buffers under a uniform flow of independently arriving cells. Buffer-states at the moment of arrival of excess-rate cells are analyzed. Formulae are given for the probabilities of changes between these events, and are deduced from these equivalent buffer/server models for use in the simulation. The second regime is that of bursty sources: classical call models are used to provide the buffer/server models for simulation.
The accelerated simulation results are compared with both full cell level simulation and, for the bursty regime, with the results of laboratory network measurements.