DIFFUSION CELL LOSS ESTIMATE FOR ATM WITH MULTICLASS BURSTY TRAFFIC

We describe a diffusion approximation model for an ATM statistical mul tiplexer using the instantaneous return model approach(1). Two Cell Lo ss Estimates are proposed for multiclass traffic. Our aim is to provid e a novel conservative, accurate and computationally efficient method for predicting cell loss probabilities which we call the Finite Buffer Diffusion Cell Loss Estimate (FBDCLE) and Infinite Buffer Diffusion C ell Loss Estimate (IBDCLE). We evaluate their accuracy by comparing th em with simulation results using a wide variety of input traffic chara cteristics. In particular we test the model with traffic which is a mi xture of different 'On-Off' sources with varying loads. Both homogeneo us and heterogeneous aggregated arrival processes have been taken into account. These comparisons, which include evaluations of the statisti cal confidence of the simulation runs, show that our model predictions are very close to the simulation results. In particular, FBDCLE is a conservative upper bound to cell loss ratio, while the other (IBDCLE) provides an accurate predictor which may slightly under-estimate or ov er-estimate cell loss.