Skip to Main Content
A large body of networking research has addressed the question of predicting steady state quality of service measures such as delay distributions and loss probabilities for various purposes such as network design (i.e., dimensioning links, buffers, switches, etc.) and admission control (making a decision whether or not to admit a connection or set of connections into the network). In a high-speed network environment the ratio of the propagation delay to the transmission delay can be very high, and by the time the control packet reaches the source, millions of packets may already be in transit and contributing to significant losses and further congestion in the network. To address this problem, the thesis currently investigating techniques of predicting the network traffic sufficiently far into the future (more than the round trip propagation delay) and controlling the non-real-time traffic based on the predicted value. The proposed idea is that a predictive/control based solution must explicitly take into account the queuing behavior at the node(s) of interest. This approach is directly in contrast with traditional predictive flow control schemes which have tried to minimize the prediction error of the source by largely ignoring the underlying queuing structure because it was too difficult to handle. The scheme of the thesis is try to minimize the congestion in the queue rather than focus on minimizing an adhoc criterion such as the mean squared error of the source. The basic system is based on the predicted value of the aggregate traffic at the bottleneck node, a control vector is generated for the non-real-time traffic sources.