Elsevier

Computer Communications

Volume 31, Issue 14, 5 September 2008, Pages 3344-3359
Computer Communications

Design of rate-based controllers for active queue management in TCP/IP networks

https://doi.org/10.1016/j.comcom.2008.05.032Get rights and content

Abstract

In our previous work [J. Aweya, M. Ouellette, D. Montuno, K. Felske, Rate-based proportional-integral control scheme for active queue management, International Journal of Network Management, John Wiley & Sons, Ltd., vol. 16, issue 3, May–June 2006, pp. 203–231] we argued that rate-based active queue management (AQM) schemes are most appropriate for high speed links which, typically, have small buffers in relation to the bandwidth-delay product of the link. We continue our discussion in this paper to study the performance of other rate-based AQM controllers. We consider integral (I) and proportional (P) rate-based controllers for the AQM problem, we then characterize all stabilizing feedback gains for the closed-loop TCP/AQM system. Using a network topology with short- and long-lived TCP flows, we present simulation results for the rate-based AQM controllers. We observe that the I-controller is able to control properly the system compared to the P-controller. The P-controller in most cases produces offsets from the target values. To further support our observations, we use a closed-loop control model and control theory to explain why the P-controller has major limitations in the control of a first-order plant like our TCP/AQM first-order plant.

Section snippets

Summary of our previous work

In our previous paper [1] we described an AQM scheme which uses traffic rate information for congestion control. Based on a non-linear fluid-flow model of TCP, we designed a proportional-integral (PI) controller for AQM. Starting with a linearized model of the non-linear TCP plant [2], [3], we derived a first-order linear plant model with time delay. We then discussed the problem of stabilizing the first-order system with time delay using a proportional-integral (PI) controller. Due to the

Dynamic model of TCP/AQM

In [2], [3], a rate-based fluid-flow model of a TCP source i accessing a single link is described by the following non-linear differential equation:r˙i(t)=1di2(t)-θri(t)ri(t-di(t))p(t-di(t)),where r˙ denotes the time-derivative of r, ri is rate of the TCP source (in packets per second), di is the round-trip time (RTT) of the TCP source (in s), p is the marking/dropping probability of a TCP packet at the link, and θ is a parameter characteristic of the type of TCP source. Similar to the model in

Simulation results

The objective of our simulation studies is to investigate how well the I controller and P-controller can maintain small queue sizes and target link utilization. The simulation results presented here were obtained using the OPNET Modeler simulation tool. Fig. 1 shows a simple bottleneck network configuration with two routers and a number of TCP flows which is the same configuration used in [1].

The routers are interconnected by a T3 (45 Mbps) link having a propagation delay of 1 ms. There are 10

Results for the integral controller

This section presents simulation results for the I-controlled TCP/AQM plant. We also discuss the properties of I-control of a first-order plant from a control theoretic perspective.

Results for the proportional controller

This section presents simulation results for the P-controlled TCP/AQM plant. Here we also discuss the properties of P-control of a first-order plant from a control theoretic perspective.

Remarks on PI control of a first-order system

A rate-based PI-controller of TCP/AQM system was described in our previous work [5]. Here we present control theoretic insights into why the PI-controller performed very well in the control the first-order plant. The PI-controller can be expressed asCAQM(s)=kp+kis,where kp and ki are proportional and integral gains, respectively. For the closed-loop system in Fig. 9, we get the following transfer functionsY(s)X(s)=kps+kias2+(b+kp)s+ki,Y(s)D(s)=-sas2+(b+kp)s+ki.The system is stable for positive

Conclusion

We have shown through simulations that properly designed I and P rate-based AQM scheme can maintain small queue sizes in the presence of long-lived FTP sources and short-lived web sources. Results were presented for networks with asymmetrical round-trip times, different link capacities and packets sizes. Maintaining small queue sizes is an attractive property for switches and routers that have very small buffers (typically much smaller than the bandwidth-delay product of the network). It is

References (10)

  • J. Aweya, M. Ouellette, D. Montuno, K. Felske, Rate-based proportional-integral control scheme for active queue...
  • S. Kunniyur, R. Srikant, End-to-end congestion control: utility functions, random losses and ECN marks, in: Proceedings...
  • S.S. Kunniyur et al.

    An adaptive virtual queue (AVQ) algorithm for active queue management

    IEEE/ACM Transactions on Networking

    (2004)
  • G.J. Silva, A. Datta, S.P. Bhattacharyya, Stabilization of time delay systems, in: Proceedings of the American Control...
  • G.J. Silva et al.

    New results on the synthesis of PID controllers

    IEEE Transactions on Automatic Control

    (2002)
There are more references available in the full text version of this article.

Cited by (16)

View all citing articles on Scopus
View full text