Congestion control in ATM networks using learning algorithms

This paper describes a 'real time' solution to the link-by-link call admission control (AC) problem in ATM networks for bursty and variable bit rate video traffic and for mixes of them. The proposed method employs SELA, a novel Stochastic Estimator Learning Algorithm, for predicting whether a new call should be accepted or not. Call acceptance decision is derived from the independent two-call and cell-level execution of two distinct learning automata whose selected actions are combined via an AND function. The feedback which the algorithms receive has been drawn from efficient 'equivalent bandwidth' approximations and accurate cell loss probability estimations. This AC mechanism exhibits a remarkable gain obtained from statistical multiplexing, compared with other schemes reported in the literature.