Résumé: Although neural networks (NNs) have been applied for traffic and congestion control in ATM networks, many implementations only consider a single service class with identical traffic characteristics and requirements. In addition, the most commonly used NN is the multi-layer perceptron (MLP) networks which are known to converge slowly. We present a multi-service connection admission control (CAC) scheme that harness the fast learning capability of modular neural networks for cell-loss-ratio (CLR) prediction at each switch. The proposed scheme does not require precise models of network processes and detailed knowledge of higher-order moments of the traffic. Simulations were carried out to compare the performance of the proposed scheme with an equivalent bandwidth (EBW) method. (c) 2001 Elsevier Science B.V.

Langue: Anglais