Simulation of ABR policers

The simulation experiments described in this paper are in three stages. The first stage is called the `worst case´; here we analyse the performance of a N-store policer, using N=2, 3, 5, 12, on rates with a uniform distribution between a pre-defined lowest and highest value. However, it is true to assume that rates are not updated every 32 cell times, but in fact are likely to occur as random clusters within a background pattern of repeated rates (i.e. having the same value within successive backward RM cells (BRMs)); so the next stage of the experiment, `approximations´, analysed the N-store policer performance using this type of traffic pattern. This stage of the experiment also allowed us to recommend an optimum size for N (i.e. beyond which gains made were insignificant). Finally, the third stage of the experiments, `bandwidth savings´, analysed the bandwidth savings which could be achieved by using the optimum sized 10 store policer, instead of the 2 store policer