Combined reactive/preventive approach for congestion control in ATM networks

An adaptive rate based congestion control mechanism based on the leaky bucket scheme is proposed. The parameters of the mechanism are controlled using (delayed) feedback information sent by the network. The network is modeled as a multiplexer fed by a number of sources each of which is controlled by a leaky bucket. A simple threshold policy at the multiplexer queue to detect congestion is examined. A Markovian state-space representation of the system is provided, which is used for an efficient simulation of the system. The effects of propagation delay and the parameters of the control scheme on the performance are studied. Results indicate that, even for large propagation delay, both an appropriate feedback policy at the source and a proper adaptive mechanism at the leaky buckets improve the system performance considerably