A control-theoretic design of RED for improved congestion control in TCP/IP routers

This paper presents a new design of the Random Early Detection (RED) router algorithm, based on a control theoretic analysis of Active Queue Management (AQM) in TCP/IP networks. General design equations are developed that can be used under a wide range of network traffic conditions. It is shown, by discrete event simulations using the ns-2 tool, that the newly designed controller has a faster response to changing traffic loads than do existing algorithms and that it yields smaller queue fluctuations. An adaptive version of algorithm that tunes itself to the prevailing traffic load is also developed.