Flow control for ABR service in ATM networks using modular neural networks

Most congestion control schemes only react to congestion after it has taken place, which may be too late in high-speed ATM networks. In this paper, we present an enhanced version of the 'intelligent congestion control scheme' by Siu & Tzeng(1) using modular and hierarchical neural networks for predicting the congestion status of an ATM switch. The fast learning and accurate predictions obtained with this architecture is shown to produce small average queue length while maintaining a high throughput. Experiments which compared the performance of this enhanced scheme with the original scheme show that the ability to predict congestion becomes increasingly important when link distances are longer.