Analysis of neural-network-based congestion control algorithms for ATM networks

Congestion control provides a challenge in the design of Asynchronous Trans fer Mode (ATM) networks. Several algorithms have been proposed in the liter ature for alleviating or reducing congestion. In this paper the Jumping Win dow (JW), Triggered Jumping Window (TJW) and the Exponentially Weighted Mov ing Average (EWMA) window algorithms are analyzed, based on a closed-loop p redictive feedback mechanism using a neural network. Single- and multiple-s ource models with real-world and simulated data are used to test the perfor mance of the proposed mechanisms. The consequences of delayed feedback mess ages is also considered. Results indicate that neural controllers are effec tive in reducing the cell loss rate, while introducing minimal additional d elays. (C) 1999 Elsevier Science Ltd. All rights reserved.