Abstract
This paper describes the development of an adaptive algorithm for active queue management (AQM) routers supporting a dynamically varying number of long-lived TCP flows. We use a nonlinear dynamic model of TCP behavior which was recently developed from fluid-flow and stochastic differential equation analysis. We first show that the high-frequency gain of the linearized plant model is inversely proportional to the network load factor which can easily vary by an order of magnitude. The motivation for adaptation follows from the direct relation between the plant’s high-frequency gain and stability margins and speed of response. Our adaptive algorithm — the Externally Excited Adaptive Loop (EEAL) — is shown to achieve improved queue level management compared with that achieved using RED AQM and the recently proposed PI AQM.
