Adaptive congestion control in ATM based networks: Quality of service and high utilisation

To improve network utilisation and robustness for Asynchronous Transfer Mode (ATM) networks there is increasing evidence that reactive congestion controls are necessary. However the large bandwidth-delay product and the difficulty in traffic modelling inherent in ATM networks makes design of effective reactive controls difficult. In this paper, we use adaptive (feedback and feedforward) predictive control techniques, and despite the (large) propagation delays avoid congestion at high server utilisation while maintaining quality of service at prescribed levels. Our method overcomes the difficulty of the large propagation delays by using adaptive traffic prediction calculated from measurements of the net input traffic at an ATM buffer. Effective prediction is possible due to the high level of correlation present in some VBR traffic. The sensitivity to traffic modelling is addressed by using adaptive (predictive) control. Bounds on operating conditions are derived analytically, and evaluated using simulation. A comparative evaluation is also presented.