In this paper we develop a new predictive flow control scheme and
analyze its performance. This scheme controls the non-real-time
traffic based on predicting the real-time traffic. The goal of the
work is to operate the network in a low congestion, high throughput
regime. We provide a rigorous analysis of the performance of our flow
control method and show that the algorithm has attractive and useful
properties. From our analysis we obtain an explicit condition that
gives us design guidelines on how to choose a predictor. We learn that
it is especially important to take the queueing effect into account in
developing the predictor. We also provide numerical results comparing
different predictors that use varying degrees of information from
the network.
|