Fixed and adaptive model-based controllers for active queue management

Abstract

In this paper we present new model-based controllers for active queue management (AQM) supporting TCP flows. Our control design is based on linearization of a previously developed TCP model using fluid flow and stochastic differential equations. An AQM using a simple PI controller is shown to significantly outperform RED AQM. However, we show that both AQMs are unable to maintain performance as the number of flows increases. To overcome this, we introduce a special type of adaptation, the externally excited adaptive loop (EEAL) controller. In comparison to the PI AQM, the EEAL AQM exhibits faster response to flow changes, and is able to give a dynamic estimate of the number of flows.