Application of model-free LQG subspace predictive control to TCP congestion control

Abstract

We investigate the application of a model-free linear quadratic Gaussian (LQG) subspace-based predictive controller to Internet congestion control. Specifically, we consider a classically designed LQG linear congestion controller with a non-standard performance index and determine whether a model-free controller is a viable alternative in this instance. We employ the model-free subspace predictive controller methodology which we customize for end-to-end transmission control protocol (TCP) congestion control. A series of network simulations support the use of the more easily implementable model-free controller over its classical analogue. We further demonstrate that the model-free controller provides increased stability under transient network conditions when compared with the first feedback congestion controller, TCP Vegas. Copyright © 2007 John Wiley & Sons, Ltd.