On a novel framework for optimizing QoS of long range dependence traffic in ATM

With the advent of high-speed networks for data and multimedia communications, traffic is becoming more and more bursty. Self-similarity, which is introduced by new data networks, is a manifestation of heavy correlation between cell/packet inter-arrival times. We examine the effectiveness of a congestion control framework in reducing the inherent variability of fractal traffic, with the objective of increasing utilization of the links and reducing the cell loss ratio in ATM networks. We propose to do so at the source by controlling the injection of the burst of cells at the user-network interface (UNI) source nodes. We show that the proposed algorithm also reduces the cell delay at individual nodes of the ATM backbone. End-to-end delay (a QoS parameter) is also shown to be reduced by the proposed algorithm. We further show that including the burst control approach reduces the buffer requirement in a congested environment to maintain the negotiated QoS