The trade-off between robustness and disturbance rejection for congestion control algorithms based on a modified smith-predictor

Congestion control is a cornerstone component of the Internet. The plant dynamics can be modelled by means of an integrator, modelling a bottleneck queue, a time delay, modelling the propagation of the information from a source to a destination along with queuing, and a load disturbance, which models the time-varying available bandwidth. It has been shown that a Smith predictor plus a proportional gain is an effective controller, even though, it is not able to reject the load disturbance. To overcome this issue, which is particularly relevant in the case of multimedia delay-sensitive traffic, we consider the modified Smith predictor proposed by Matausek and Micic as a candidate for the design of a congestion control algorithm. By taking a geometric approach, we quantify the trade-off between disturbance rejection property of the modified Smith predictor and the achievable stability robustness with respect to delay uncertainty, which in data networks is due to queuing. Finally, we propose some guidelines to tune the additional parameter introduced by the modified Smith predictor.