Elsevier

Computer Communications

Volume 30, Issue 8, 8 June 2007, Pages 1731-1741
Computer Communications

RaQ: A robust active queue management scheme based on rate and queue length

https://doi.org/10.1016/j.comcom.2007.02.005Get rights and content

Abstract

This paper introduces a new Active Queue Management (AQM) scheme, which is both rate-based and queue-based called RaQ and presents its design details. We present a stability analysis for a TCP/RaQ system that leads to a sufficient condition for the system to be stable. The performance of RaQ is evaluated by simulations and the results demonstrate that RaQ AQM is stable and robust under various scenarios including cases involving non-TCP traffic types (UDP and HTTP) and a case involves a multiple bottleneck topology. Its superiority over other AQMs is also demonstrated.

Introduction

Active Queue Management schemes have been proposed to complement the TCP network congestion control. AQM is a router-based congestion control mechanism which aims to signal congestion early by dropping packets before the buffer becomes completely full. The ability of AQM to convey congestion notification to the end-hosts enables the sources to reduce their sending rates prior to buffer overflow. Explicit Congestion Notification (ECN) proposed in [17] allows an AQM to notify sources of congestion using packet marking instead of dropping packets.

AQM schemes are driven by the following goals: low buffer occupancy resulting in small queueing delays, low queue length jitter, low packets losses and high link utilization. Preferably, an AQM scheme would make a good choice of packet dropping/marking at each congestion stage (measured by queue length for example) so that the above mentioned goals are realized. The second desirable feature of a good AQM scheme is its robustness. A robust AQM scheme would require little tuning by a network operator and would remain inherently stable to traffic fluctuations [12].

Many AQM schemes have been studied in recent literatures. These schemes can be classified into three types. The first includes AQM schemes that are based on queue length measurements. This category is also called queue-based AQM scheme, which uses the average (or instantaneous) queue length to compute the packet dropping/marking probability. Random Early Detection (RED) [6], its modification [3], [7], [19], and PI control algorithm [10], Dynamic RED [2], PD-Controller [20], belong to this category. The second group includes AQM schemes that use input traffic rate measurements. These are called rate-based AQM schemes. Blue [4], GREEN [23] and adaptive virtual queue (AVQ) [9] are rate-based AQM schemes. The third group includes AQM schemes that use both queue length and rate. The instantaneous queue length and input traffic rate are both deployed to determine the packet dropping/marking probability in the category. Random Exponential Marking (REM) [1], Virtual Rate Control (VRC) [16] and Yellow [12] are representative algorithms.

This paper presents a new integrated queue-based and rate-based AQM scheme called RaQ. RaQ uses the input rate and current queue length to calculate the packet dropping/marking probability. From the point of control theory, RaQ can be seen as dual loop feedback control. The inner loop is rate feedback control and the outer loop is queue length feedback control. Thus, the rate feedback control enables RaQ to respond to congestion rapidly, so it can decrease the packet loss due buffer overflow, the queue length feedback control stabilizes RaQ’s queue length around given target, so it can achieve predictable queueing delay and lower delay jitter.

The rest of the paper is organized as follows. Section 2 describes the RaQ AQM in detail, providing a stability analysis, and choices of parameters. Section 3 provides background details of several well-known AQM schemes. Section 4 presents performance evaluation of RaQ for single bottleneck network and Section 5 presents performance evaluation of RaQ for multiple bottleneck network. Section 6 presents comparison of RaQ with the other AQM schemes described in Section 3.

Section snippets

Description

RaQ is an integrated rate-based and queue-based AQM scheme. It uses Proportional rate control and Proportional–Integral queue length control as follows:p(t)=Ψ{rkp(r(t)-C)+qkp(q(t)-QT)+qki0t(q(t)-QT)dt}where p(t) is the packet dropping/marking probability, r(t) is the aggregate input rate of the queue, C is the link capacity, q(t) is the instantaneous queue length, QT is the target queue length, rkp is the Proportional coefficient of rate control, qkp and qki are the Proportional and Integral

Related work

The traditional queue management scheme is Drop-Tail. Only when the buffer is full, packets are dropped by Drop-Tail. Thus, Drop-Tail has the problems of lock-out, global synchronization, and full queues. To overcome the drawbacks of Drop-Tail, AQM has been proposed. The basic rationale of AQM is to provide advance notification of congestion to the sender, so that the sender can reduce its transmission rate before buffer overflow occurs. One of the most prevalent AQM algorithms is RED [6]. RED

Performance of RaQ: the single bottleneck case

We study the performance of RaQ via simulations. In our simulation testing we focus on stabilizing queue length at a target value QT as a key performance measure. As discussed, if we can control the queue to stay close to a desirable target, we can achieve high throughput, predictable delay and low jitter. The low jitter enables meeting Quality of Service (QoS) requirements for real time services especially when the queue length target is achieved independently of traffic conditions [24].

Performance of RaQ: the multiple bottleneck case

Here we extend the simple single bottleneck topology to a case of multiple bottlenecks. We consider the network topology presented in Fig. 13. There are two bottlenecks in this topology. One is between Router B and Router C, and the other is between Router D and Router E. The link capacity of the two bottlenecks is 45 Mb/s and the capacity of other links is 100 Mb/s. There are three traffic group. The first group has N TCP connections traversing all links, the second group has N1 TCP connections

Comparison with other AQMs

In this section, we perform a simulation to compare the performance of RaQ with REM [1], PI controller [10], and ARED [7]. The network topology used in the simulation is the same as in Fig. 2. The same parameters as in Section 4 are used: packet size is 1000 bytes, common link capacity is 45 Mb/s, the round-trip propagation delay of the TCP connections is uniformly distributed between 50 and 500 ms, the buffer size is 1125 packets. The target queue length is set at 300 packets for all AQM schemes.

Conclusions

We have developed a novel AQM scheme called RaQ based on the integrated input rate and queue length. From the point of control theory, it is a dual loop feedback control. The stability of RaQ is analyzed and a sufficient condition is given to guarantee the stability of the system. We have demonstrated by simulations that RaQ is able to maintain the queue length around the given target under different traffic loads, different RTPTs, and different bottleneck link capacities. Further simulation

Acknowledgements

The work described in this paper was jointly supported by grants from the Australian Research Council (Grant DP0559131), and the Natural Science Foundation of Jiangsu Province, China (No. BK2004132).

Jinsheng Sun was born in Jilin, China. He received the B.S., M.S. and Ph.D. degrees in Control Science from Nanjing University of Science and Technology in 1990, 1992 and 1995, respectively. Since 1995, he has been with the Department of Automation, Nanjing University of Science and Technology (NUST). Currently, he is on a two-year leave from NUST to take up a Research Fellow position at the Department of Electrical and Electronic Engineering, The University of Melbourne. His research interests

References (23)

  • C.V. Hollot et al.

    A control theoretic analysis of RED

    Proc. IEEE INFOCOM

    (2001)
  • Cited by (0)

    Jinsheng Sun was born in Jilin, China. He received the B.S., M.S. and Ph.D. degrees in Control Science from Nanjing University of Science and Technology in 1990, 1992 and 1995, respectively. Since 1995, he has been with the Department of Automation, Nanjing University of Science and Technology (NUST). Currently, he is on a two-year leave from NUST to take up a Research Fellow position at the Department of Electrical and Electronic Engineering, The University of Melbourne. His research interests include congestion control and fault-tolerant control.

    Moshe Zukerman received his B.Sc. in industrial engineering and management and his M.Sc. in operations research from Technion - Israel Institute of Technology and a Ph.D. degree in Electrical Engineering from the University of California Los Angeles in 1985. Dr. Zukerman was an independent consultant with IRI Corporation and a post-doctoral fellow at UCLA during 1985–1986. During 1986–1997 he served in Telstra Research Laboratories (TRL), first as a research engineer and, during 1988–1997, as a project leader. In 1997 he joined The University of Melbourne where he is now a professor responsible for promoting and expanding telecommunications research and teaching in the Electrical and Electronic Engineering Department. He has also taught and supervised graduate students at Monash University during 1990–2001. He served on the editorial board of the Australian Telecommunications Research Journal, Computer Networks, and the IEEE Communications Magazine. He also served as a Guest Editor of IEEE JSAC for two issues: Presently, he is serving on the editorial board of the IEEE/ACM Transactions on Networking, IEEE Journal of Selected Areas in Communications and the International Journal of Communication Systems. Professor Zukerman is a Fellow of the IEEE.

    View full text