Abstract

The feasibility of a proactive congestion control approach for broadband networks, especially involving large propagation delays is examined. The method is based on network overload prediction from a current underload state, by exploitation of the traffic correlations that services being transported exhibit, and subsequent shaping of input traffic through a tunable leaky bucket technique. First passage time analysis for Markov processes and Markov drift processes is used to give a measure of congestion imminence, through the probability distribution for the time to overload. Applications both to real time services and delay tolerant services are discussed. The method is predicated on knowledge of the current source state and a method of estimating it from network measurements is also presented.