RATE REGULATION WITH FEEDBACK CONTROLLER IN ATM NETWORKS - A NEURAL-NETWORK APPROACH

In this paper, we propose the use of an artificial neural network (ANN ) technique for a rate-based feedback controller in asynchronous trans fer mode (ATR-I) networks, A leaky bucket (LB) mechanism is used to do cell. discarding, when the traffic violates a predefined threshold. S ince the network cannot rely on the user's compliance with its declare d parameters, it is extremely difficult to select the best threshold v alue and depletion rate for the LB, We propose an ANN model which moni tors the status of the LB and predicts the possible cell discarding at the LB in the near future. The source rate is regulated to a certain amount depending on the feedback signal ''strength'' when possible cel l discarding is detected. The lower the value carried in the feedback cell, the higher the possibility of cell discarding and, subsequently, the higher the probability that the traffic is regulated to a lower r ate. Our model considers the propagation delay time of the feedback si gnal making our approach more realistic This mechanism is transparent to the source if the LB is correctly set up and the traffic follows it s declared parameters, We use the same trained ANN for different MPEG traces and the results of a simulation study suggest that our mechanis ms provide simple and effective traffic management for ATM networks. C ell loss rate due to the congestion shows a two to five times improvem ent compared with the static approach, while transmission delays intro duced by our ANN controller are also smaller than in the static approa ch, Channel utilization is also improved, showing that our mechanisms provides a better alternative to static feedback controllers.