The application of fuzzy prediction for the improvement of QoS performance

Abstract

A fuzzy logic based scheme aimed at the provision and maintenance of optimum queue length for the ABR class buffer is proposed and analysed. The investigation is focused under conditions of heavy offered load and long propagation delays which exist in MAN and WAN. A fuzzy logic predictor is proposed for an ATM switch to estimate the output queue length. This information together with the current queue length and growth rate is provided to a fuzzy inference system that generates an additional traffic rate factor. This factor can be used alone to increase/decrease ABR source delivery rate, or in conjunction with other congestion avoidance and control algorithms such as explicit rate indication for congestion avoidance (ERICA) to calculate ABR traffic bandwidth allocation and the explicit rate (ER) field in backward RM (BRM) cells. Simulation results indicate that this scheme improves the QoS performance of ABR and other real-time service classes. Since the fuzzy systems involved are all instantaneously functional, there is no need for lengthy training process like some neural network systems and the scheme is not affected by different classes and class combinations of traffic. With the implementation of this algorithm, both the efficiency and QoS can be improved for heavily loaded metropolitan/wide area network.