An Improved PI Active Queue Management Algorithm Based on Network State Parameters Estimation

As an important active queue management scheme, PI controller eliminates the steady state error of queue length with the introduction of integral factor, improving throughput while reducing queuing delay. But it can not adapt its control parameters when network state varies with time. So when traffic flows change, the PI controller can not converge quickly with the traffic flows. Based on the thoughts of detecting and estimating network state information through the network flows, the amount of active flows, average round trip time (RTT) and capacity of outgoing link are estimated. The amount of active flows is estimated by calculating the inverse proportion of hit function probability. The network capacity is estimated by average data packets per unit time. The average RTT is estimated by the relation equation of the amount of active flows, network capacity and data packets loss probability in the steady state. A new scheme called fast convergent PI (FCPI) is proposed based on TCP-AQM system model and the network state parameters estimation, in which the controller can be adjusted according to the real-time network state. Simulation results show that the new scheme not only improves the convergent rate of PI controller, but also appears to be robust and effective in different scenarios, which makes it more suitable for high speed routers.