A control theoretical approach to congestion control of TCP/AQM networks

This paper shows how control techniques, such as PID or generalized predictive control, can improve the performance of TCP/IP networks when dealing with congestion. Drop tail, or more sophisticated AQM techniques such as RED, perform worse than process control techniques. A comparison between the different methods in different working scenarios is analyzed. Two different GPC controllers are evaluated. One considers the model resulting from on-line identification and the other is the linearization of the non-linear model.