Adaptive fuzzy traffic predictor and its applications in ATM networks

Traffic prediction is a new research subject in ATM traffic management. It is the basis of dynamic and adaptive traffic control. This paper, based on the theory of fuzzy system, presents a fuzzy traffic predictor using nearest neighborhood clustering learning algorithm. We examine its prediction effect through autoregressive (AR) process and real video traffic. The results show that the approximator/predictor is accurate and flexible. The applications of the traffic predictor in ATM networks include dynamic bandwidth allocation, traffic shaping, traffic smoothing, service synchronization etc. We mainly discuss dynamic bandwidth allocation based on the fuzzy traffic prediction. The research results show that the dynamic bandwidth allocation can reduce cell loss ratio, buffer requirement and improve the bandwidth utilization remarkably