A predictive bandwidth management scheme and network architecture for real-time VER traffic

Many bandwidth management schemes for ATM network require detail traffic description and accurate traffic model. These informations, however, are not always available, especially in the case of real-time VER video traffic. Predictive bandwidth management scheme solves this problem by using the on-line traffic measurement-it predicts the future traffic rate from the measurement and allocates the bandwidth accordingly. In this article, we firstly introduce an adaptive wavelet predictor for dynamic bandwidth allocation. Our simulation results show that, compared with the time-domain least-mean-square (LMS) predictor, our wavelet predictor improves the prediction accuracy and significantly reduces the cell-loss-rate when used in the dynamic bandwidth allocation. We secondly present a prediction-based bandwidth allocation scheme and the corresponding network architecture. In particular, a Bandwidth Allocation Unit (BAU) at the network access node is suggested. (C) 1999 Elsevier Science B.V, All rights reserved.