The use of learning algorithms in ATM networks call admission control problem: a methodology

Abstract

The call admission control (CAC) problem, one of the most fundamental in ATM networks, has yet to be solved. We use a novel stochastic estimator learning algorithm (SELA) to predict in "real time" if a call request should be accepted or not, for various types of traffic sources. The feedback the algorithm receives has been drawn from the efficient "equivalent bandwidth" approximation proposed by Guerin et al. (1991). The proposed scheme exhibits a remarkable statistical gain compared with other CAC schemes reported in the literature, without QOS deterioration. The paper contains a simulation study of its performance and discusses several possible ways in which this work could be extended.