Technique for TCP over ATM traffic prediction in the presence of traffic policing

ATM (asynchronous transfer mode) traffic management techniques require tools that can predict the value of traffic. This paper proposes a new method of TCP self-similar traffic prediction. A self-similar traffic is characterised by an autocorrelation function that decays as a power of the lag time. Based on computer simulation results, it is concluded that the multilayer neural network is a useful tool for self-similar traffic prediction.