Performance evaluation of bursty traffic using neural networks

Abstract

We investigate the application of neural networks to evaluate the performance, packet loss probability, of a bursty traffic stream. We show, that in a bursty multimedia environment, performance is a function of burstiness, Hurst parameter, traffic intensity and buffer size. In a closed loop traffic control system each source uses this reported measure to regulate their traffic to the destination queue. A multilayer neural network is used to capture the functional relationship between the loss probability and the traffic descriptor (Hurst parameter and traffic intensity) for a fixed value of buffer size. The neural network approach makes practical real-time performance measurement and hence the control of traffic in an adaptive environment.