Estimation and prediction of VBR traffic in high-speed networks using LMS filters

portion of the traffic carried by high speed packet-switched broadband integrated services digital networks (B-ISDN). The high peak-to-mean ratio of VBR traffic, however, makes allocation of network resources, such as bandwidth, quite difficult. If bandwidth is allocated based on the peak rate, it will be highly inefficient and would, in fact, mitigate the gains of statistical multiplexing. On the other hand, choosing the mean rate means of allocating bandwidth would lead to congestion in the network, specially if multiple sources become active at the same time and start delivering video traffic into the network at their peak rates. The dynamics of the VBR traffic, coupled with the delay-bandwidth product of the high-speed networks, add to the complexity of congestion control and render conventional reactive based congestion control schemes as highly inefficient. Bandwidth allocation, based on the mean rate aided by a preventative congestion control scheme that could predict the onset of congestion, well in advance, is the desired solution. One such technique is proposed that estimates the current traffic state, predicts the future traffic states and forecasts the occurrence of congestion in the network. The emphasis is on demonstrating the applicability of least mean square (LMS) filters in estimating the current state of the VBR traffic sources accessing the networks. Some simulation results are presented.