Neural networks based variable bit rate traffic prediction for traffic control using multiple leaky bucket

Abstract

This work presents a novel feedback rate regulator using the multiple leaky bucket (MLB) for variable bit rate (VBR) self-similar traffic that is based on the traffic load prediction by time-delayed neural networks in ATM networks. In the MLB mechanism, the leak rate and buffer capacity of each leaky bucket (LB) can be dynamically adjusted based on the buffer occupancy. A finite-duration impulse response (FIR) multilayer neural network is used to predict the incoming traffic load and pass the information to the feedback rate regulator. Ten real world MPEG1 and ten synthesized traffic traces are used to validate the performance of the MLB and the MLB with an FIR prediction mechanism. Simulation results demonstrate that the cell loss rate using MLB and MLB with an FIR filter-based predictor can be significantly reduced compare to the conventional leaky bucket method.