Estimation and prediction approach to congestion control in ATM networks

An estimation-prediction approach to congestion control in ATM networks is proposed. The method attempts to achieve efficient utilization of available bandwidth by taking preventive measures long before a congestion occurs. The estimation and the prediction of the traffic model is based on the fact that the underlying traffic model is a Markov process. Cell counts in consecutive fixed frames are observed for the purpose of estimating the sources state. Kalman filtering, is used to obtain estimates. The first passage time from the present state to an overload is treated using matrix spectral expansion. Numerical results illustrating the technique are given