Cost of AQM in stabilizing TCP

Abstract

In this paper, we propose a unified mathematical framework based on receding horizon control for analyzing and designing AQM (Active Queue Management) algorithms in stabilizing TCP (Transfer Control Protocol). The proposed framework is based on a dynamical system of the given TCP and a linear quadratic cost on transients in queue length and flow rates. We derive the optimal receding horizon AQMs (RHAs) that stabilizes the linearized dynamical system with the minimum cost. Conversely, we show that any AQM with an appropriate structure solves the same optimal control problem with appropriate weighting matrix. We interpret existing AQM's such as RED, REM, PI and AVQ as different approximations of the optimal AQM, and discuss the impact of these approximations on performance.