Feedback congestion controller for ATM networks using a neural network traffic predictor

One of the fundamental challenges facing broadband information transport is to determine congestion control strategies to support multiple classes of traffic in the asynchronous transfer mode (ATM) based networks. Monitoring the buffer status is the most commonly used mechanism to detect congestions in ATM networks. However, in static feedback controllers defining the threshold of the buffer for congestion is not so direct and the degree of source rates to be regulated is not so clear, either. In this paper, we propose an explicit congestion mechanism for ATM networks using an artificial neural network to predict the traffic arrival patterns. The predicted data rate in conjunction with the current queue information of the buffer is used to generate a value that will inform the source to reduce its transmission rate. The results of a simulation study are presented which suggest that our mechanism provides a simple and effective traffic management for ATM networks. Cell loss due to congestion shows a 5 to 10 times improvement compared with the static approach. Transmission delay of our ANN controller is also smaller.