Congestion Control in TCP/AQM Networks Using a Disturbance Observer

Abstract

This paper presents a novel congestion controller for transmission control protocol/active queue management (TCP/AQM) networks. In order to maintain the queue length to a value less than the buffer size at the aggregation nodes, the queue management mechanism drops some packets depending on the probability calculated by a congestion controller. However, conventional controllers such as random early detection (RED) algorithms and proportional-integral-derivative (PID)-based controllers are not compatible with parameter variation in TCP/AQM networks. Parameter variation in TCP/AQM networks includes the change of the number of TCP connections, the inflow of user datagram protocol (UDP) flows and so on. In the proposed method, the effect of parameter variation is estimated as drop probability disturbance by using a disturbance observer (DOB). By using the DOB, disturbance suppression characteristics of the controlled system can be drastically improved. In addition, the TCP/AQM network system is analyzed on the basis of an acceleration control scheme. The proposed method is validated by simulation studies performed using a nonlinear model of a TCP/AQM network and verifications made using the network simulator ns-2.