Application of a PID feedback control algorithm for adaptive queue management to support TCP congestion control

Abstract

Recently, many active queue management (AQM) algorithms have been proposed to address the performance degradation of end-to-end congestion control under tail-drop (TD) queue management at Internet routers. However, these AQM algorithms show performance improvement only for limited network environments, and are insensitive to dynamically changing network situations. In this paper, we propose an adaptive queue management algorithm, called PID-controller, that uses proportional- integral-derivative (PID) feedback control to remedy these weaknesses of existing AQM proposals. The PID-controller is able to detect and control congestion adaptively and proactively to dynamically changing network environments using incipient as well as current congestion indications. A simulation study over a wide range of IP traffic conditions shows that PID-controller outperforms other AQM algorithms such as Random Early Detection (RED) [3] andProportional-Integral (PI) controller [9] in terms of queue length dynamics, packet loss rates, and link utilization.

PI-PD Controller for Adaptive and Robust Active Queue Management for Internet Congestion Control

“…The PID controller [7] was designed to improve the speed of response and to eliminate (or minimize) the steady-state error of the TCP/AQM system at the same 2. Frequency response and time response methods are often called frequency domain and time domain methods.…”

“…Since the controlled system, the TCP flow dynamics (1), is a second-order system, time domain performance specifications such as the rise time, the settling time, the maximum overshoot (MOS), the time constant, and the steady-state error are available analytically [26]. Thus, the PID controller has been designed in Ryu and Rump [7] using the time domain design and analysis method.…”

“…7 These 9 pairs of TCP/Reno sources and destinations are connected to nc0 and nc1, 7. In general, Internet flows go through many routers and may experience congestion at some of those routers on the way from a source to its destination.…”

“…In the PID controller [7], we set the desired maximum overshoot to 5% (i.e., ξ = 0.6901 equivalently) and the settling time (t s ) to the time to reach and stay within ±2% of the steady-state value in the design of a PD control part for a given network configuration. We set R 0 /2 as the time constant for the design of the PD control part because it gives better control dynamics than R 0 in terms of speed of response and the steady-state error.…”

“…Hence, the congestion detection and control in these algorithms are reactive to current or past congestion, not proactive to incipient congestion [6,7]. For example, RED can detect the long-term traffic patterns using exponentially weighted average queue lengths and notify the onset (persistent) congestion to sources.…”