A subspace predictive controller for end-to-end TCP congestion control

Abstract

In this paper we present a subspace-based model-free LQG linear congestion controller for TCP, intended as a viable alternative to its classically designed model-based counterpart. The model-free controller is derived from TCP Vegas and is constructed directly from experimental data, without requiring a plant model. We show the controller possesses the same equilibrium behaviour as Vegas, while offering substantially better transient behaviour. Simulation results support the use of the more easily implementable model-free controller.