An adaptive LQG TCP congestion controller for the Internet

Abstract

This paper addresses the problem of congestion control for transmission control protocol (TCP) traffic in the Internet. The method proposed builds on the ideas of TCP Vegas, a true feedback control approach to congestion management of TCP traffic. The new method is based on an adaptive linear quadratic Gaussian (LQG) formulation which uses an extended least squares system identification algorithm com- bined with optimal LQG control. Simulation experiments indicate that the new technique inherits good equilibrium properties from TCP Vegas, but has much superior transient responses which, the paper argues, is important for good dynamic congestion control.