Analysis and Design of AQM for stabilizing TCP

In this paper, we formulate the AQM (Active Queue Management) design problem for stabilizing a given TCP (Transmission Control Protocol) as state-space models. First, we propose a PD-type (Proportional-Derivative) control structure and by applying integral control action, a PID-type (Proportional-Integral-Derivative) control structure that is a uni ed AQM framework. Second, we compensate for delays in congestion measure explicitly by using a memory control structure. Third, we propose stabilizing optimal AQMs for linearized systems of the given TCP by minimizing linear quadratic costs of the transients in queue length, aggregate rate, jitter in the aggregate rate, and the congestion measure, which is called RHA (Receding Horizon AQM) in this paper. We also show that any AQM with an appropriate structure solves the same stabilizing optimal control problem with appropriate weighting matrices. We interpret existing AQMs, including a simpli ed RED (Random Early Detection) without a low-pass lter and saturation functions, REM (Random Exponential Marking), PI (Proportional-Integral) and a simpli ed AVQwithout an adaptive virtual-queue dynamics, as di erent approximations of the uni ed AQM structure. Finally, we discuss the impact of each structure on performance from the results of the stabilizing optimal AQMs. We illustrate our results through simulation examples for the linearized system of the given TCP and queue dynamics. We submitted this paper to IEEE Conference on Decision and Control, March 1, 2002 (Paper ID: CDC02REG1058). This paper updated the technical report [1]. The corresponding author, Ki Baek Kim is visiting Caltech from Engineering Research Center, Seoul National University, Korea