Content uploaded by Kyung-Joon Park
Author content
All content in this area was uploaded by Kyung-Joon Park on May 20, 2014
Content may be subject to copyright.
A preview of the PDF is not available
Active queue management algorithm with a rate regulator
Abstract
In this paper, we propose an efficient control scheme for active queue management (AQM) supporting TCP flows. The proposed controller consists of two parts: a rate controller and a queue size controller. The rate controller is a proportional-integral (PI) controller, which improves the response to dynamic traffic variation and keeps the packet arrival rate around the link capacity. The queue size controller is a proportional (P) controller like the Random Early Detection (RED) algorithm. The rate controller gains are obtained by minimizing the performance index, either the integral square error (ISE) or the integral absolute error (IAE). We compare the performances of the proposed algorithm, the RED algorithm and the PI controller for AQM through ns simulations.
... In [13], VRC was shown to be a proportional–integral–derivative (PID) controller. Another form of PID controller [14] and a PID controller based on the state-space approach [15] have also been proposed recently. The current paper is an extended version of the work done in [12,13]. ...
... We compare the performance of VRC when the access links are heterogeneous and the different transport protocols are used and when the links are homogeneous and the same protocol is used. In the homogeneous case, we set C s and T p;s for all subnets to have same values, 50 Mb/s and111213141516171819 ms, respectively. Note that the average capacities and the average propagation delays for the homogeneous case and the heterogeneous case are nearly the same.Table 4 summarizes and compares the performance of VRC. ...
... 16. The capacity and propagation delay of each link between routers is set to T3 rate (45 Mb/s) and5678910 ms, respectively, while the links between routers and cross-traffic senders/receivers have a capacity of 100 BaseT rate (100 Mb/s) and a propagation delay of567891011121314151617181920 ms. There are N flows traversing all links and N k flows traversing each individual link k. ...
The virtual rate control (VRC) algorithm has been proposed for active queue management (AQM) in TCP networks. VRC, a rate-based control mechanism, responds quickly to traffic changes, thus allowing for high utilization and small loss. It can effectively stabilize both the input rate and the queue length around their target levels. In this paper, we analyze the stability of the VRC algorithm based on a linearized TCP model with time delay and provide a design guideline for parameter setting to make the overall system stable. Finally, we confirm the validity of our analysis and the effectiveness of VRC compared to RED, PI, REM, and AVQ through extensive ns-2 simulations.
... P-type AVQ in terms of aggregate, respectively, for closedloop stability. The papers [13], [14] have suggested PI-type AQM in terms of aggregate in an inner loop. However, all these results based on the transfer function approach do not consider what is a natural state-feedback control to stabilize the given TCP and queue dynamics. ...
... This paper attempts to address this issue by applying a modified virtual queue dynamics. For simplicity of designing a stabilizing AQM as in [10], [11], [13], [14], we assume single link and homogeneous sources with the same window, and no short flows, where the real network is asymmetric as shown in [20]. We also do not consider input/state constraints and uncertainty which are intrinsically included in real networks and ns simulator, although our work can be extended to more realistic problems using modern control theories and technologies. ...
... How to compensate for the delay explicitly and how to obtain a stabilizing gain H P is discussed later in this paper. The derived P-type AQM structure, which we also did not expect, is interesting again as follows: If we linearize the AVQ p(t) = p( ˙ ˜ b(t)) in [7], [12] , we also have a Ptype control structure δp(t) = H a P δ ˙ ˜ b(t), i.e., our study supports their argument that the AVQ is an appropriate control structure to stabilize TCP Reno based on the virtual queue dynamics in the absence of delays As shown in [8], [11], [13], [17], [18], [19], we can make the steady-state tracking error approach zero faster by adding integral control action [21] to the original system as follows: ...
This paper studies how to design a stabilizing AQM in supporting TCP (Transmission Control Protocol) with arbitrary delays. For the well-known AIMD (Additive Increase Multiplicative Decrease) dynamic model of TCP, our study shows that we can compensate for arbitrary delays explicitly by applying a modified virtual queue dynamics that makes the equilibrium queuing delay zero. This study also verifies that a simplified AQM AVQ (Adaptive Virtual Queue) is the state-feedback control for the AIMD model based on the virtual queue dynamics.
... The above state-space model is a minimal representation with state-space variables (AÉص, AÉص , AÉص ) of (4). The control strategy based on a state-space model is called modern control in the control literature, while that based on a transfer function model as in [14], [15], [16], [17], [18] is called classical control. Since the state-space approach was developed in 1950s, it has been widely investigated due to many advantages over the transfer function approach. ...
... As a sequence, papers [15], [16] have investigated how to scale gains of PI-type REM (Proportional-Integral) in terms of queue length and Ptype AVQ in terms of aggregate, respectively, for closed-loop stability. The papers [17], [18] have suggested PI-type AQM in terms of aggregate in an inner loop. However, all these results based on the transfer function approach do not consider what is a natural state-feedback control to stabilize the given TCP and queue dynamics. ...
... This work builds upon the AIMD model introduced in [13], [14]. As a first step to get possible basic results for more realistic cases, we study a simplified version of this problem, simplified in three regards as in [14], [15], [17], [18], [21]. First, we assume single link, homogeneous sources with the same window size, and no short flows (Some extension of this paper to multiple links and heterogeneous delays is discussed in our companion paper [22]). ...
In this paper, we investigate how to design AQM with a low-pass filter (average queuing) in supporting TCP based on the well-known AIMD dynamic model sharing a common bottleneck router. Since we formulate the AQM design problem for the given TCP as a state-space model, we get two interesting and important results: (1) we derive PD-type (proportional-derivative) AQM structure is derive in terms of queue length which supports the conjecture in the literature that the AQM RED is not enough to control TCP dynamic behavior, where RED can be classified as P-type AQM structure; (2) we propose a delay-dependent AQM structure to compensate for delays in congestion measure explicitly from the knowledge of RTT, capacity and number of sources, where the previous AQMs RED, REM/PI and AVQ are delay-independent controls. As one trial to obtain stabilizing gains of the proposed structures, we minimize a LQ (linear quadratic) cost of the transients on average-queue length, average-queue length rate, jitter in the rate and the loss probability. Finally, we illustrate the above theoretical results through ns simulations for TCP Reno.
... As a follow up, papers [19] and [20] have investigated how to scale gains of proportional-integral (PI)-type REM in terms of queue length and P-type AVQ in terms of aggregate, respectively. Papers [21] and [22] have suggested PI-type AQM in terms of aggregate in an inner loop and P-type AQM with a low-pass filter in terms of aggregate, respectively. Although all these papers suggest stabilizing conditions from the linearized systems, they have not focused on what kind of control structures are necessary to fully control the closed-loop system TCP/AQM fair. ...
... This paper tries to address these issues as follows: In order to get basic results for more realistic cases, we study a simplified version of this problem as in [18], [19], and [21]- [24]. First, we consider the linearized system of the AIMD and queue dynamics around the equilibrium. ...
... Let's assume that we use in (9) for the delayed system (7) [or (17) for (16)]. This kind of controls including RED, REM, PI, AVQ, and AQMs in [21] and [22] are called delay-independent (or memoryless) controls in the literature. Then, the closed-loop system is given by (or and thus has infinite number of eigenvalues. ...
This paper investigates how to design feedback controls supporting transmission control protocol (TCP) based on the state-space approach for the linearized system of the well-known additive increase multiplicative decrease (AIMD) dynamic model. We formulate the feedback control design problem as state-space models without assuming its structure in advance. Thereby, we get three results that have not been observed by previous studies on the congestion control problem. 1) In order to fully support TCP, we need a proportional-derivative (PD)-type state-feedback control structure in terms of queue length (or RTT: round trip time). This backs up the conjecture in the networking literature that the AQM RED is not enough to control TCP dynamic behavior, where RED can be classified as a P-type AQM (or as an output feedback control for the linearized AIMD model). 2) In order to fully support TCP in the presence of delays, we derive delay-dependent feedback control structures to compensate for delays explicitly under the assumption that RTT, capacity and number of sources are known, where all existing AQMs including RED, REM/PI and AVQ are delay-independent controls. 3) In an attempt to interpret different AQM structures in a unified manner rather than to compare them via simulations, we propose a PID-type mathematical framework using integral control action. As a performance index to measure the deviation of the closed-loop system from an equilibrium point, we use a linear quadratic (LQ) cost of the transients of state and control variables such as queue length, aggregate rate, jitter in the aggregate rate, and congestion measure. Stabilizing gains of the feedback control structures are obtained minimizing the LQ cost. Then, we discuss the impact of the control structure on performance using the PID-type mathematical framework. All results are extended to the case of multiple links and heterogeneous delays.
... As a consequence, when a TCP source finds out such preventive drops, it reduces the sending rate according to the additive increase multiplicative decrease (AIMD) algorithm inherent to the TCP protocol. In the recent years, various works have been done to enhance RED's performance (Floyd et al., 2001;Lim et al., 2002;Sun et al., 2003;Chen and Yang, 2009;Abbasov and Korukoglu, 2009). Some of these works are trying to increase RED's performance by improving its drop probability calculation function (Xiong et al., 2008(Xiong et al., , 2010Liu et al., 2005;Zhou et al., 2006;Cho et al., 2008) and some other works improve it by dynamic tuning of its parameters such as minth and maxth (Jamali and Zahedi, 2010;Chen and Yang, 2009). ...
... So, when the average queue becomes larger than max th , RED often does not perform well and resulting in significantly decreased throughput and increased dropping rates. This weakness has been improved in Gentle-RED (Lim et al., 2002). In Gentle-RED the packet dropping probability increases linearly from max p to 1 when average queue length increases from max th to 2nmax th (see Fig. 1b). ...
RED and most of its family algorithms use only the average queue length as a congestion meter. Since the average queue length considers only long-term behavior of the queue, these algorithms fail to see instantaneous changes of the queue length and hence their reaction to the congestion is not fast enough. In other words the feedback generated by using only the average queue length does not reflect the network congestion precisely and hence leads to a poor performance and stability. This paper solves this problem by designing a RED-based active queue management (AQM) algorithm, called FUF-RED that provides a Full Information Feedback. This algorithm not only considers the average queue length but also it takes into account growth rate of the instantaneous queue length to calculate its congestion feedback. The proposed algorithm is supported by a theoretical stability analysis which gives those feedback gains that guarantees the network stability. Extensive packet level simulations, done by using ns-2 simulator, show that the proposed algorithm outperforms existing AQM algorithms in terms of stability, average queue length, number of dropped packets and bottleneck utilization.
... Several AQM policies that use proportional-integralderivative controllers have also been proposed in the literature [18][19][20][21][22][23]. Among them, the VCR AQM policy [19] tries to stabilize both the input rate and the queue length close to their target levels by employing the notion of virtual arrival rate. ...
... Among them, the VCR AQM policy [19] tries to stabilize both the input rate and the queue length close to their target levels by employing the notion of virtual arrival rate. Moreover, in [20] a rate-based AQM policy is introduced, and in [23] a comparison of rate-based and queue-based models of congestion control is presented. ...
Several variants of the Transmission Control Protocol (TCP) have been proposed for overcoming the inefficiency of the TCP Reno when operating in high-speed networks. However, these variants by themselves cannot prevent network congestion. Congestion avoidance also relies on other mechanisms, such as Active Queue Management (AQM). This paper introduces a novel optimal AQM policy to operate in networks that adopt the High Speed TCP (HSTCP) variant as their transport protocol. The effectiveness of various optimal controllers is evaluated and the performance of the new policy is compared to that of Random Early Detection (RED) policy. Results, derived via simulation, reveal the advantages of adopting HSTCP-H2 in large bandwidth-delay product networks.
... Considerable research has been undertaken on active queue management (AQM) for congestion control in TCP networks. The Random Early Detection (RED) gateways and many other algorithms have been proposed for AQM [1] [2] [3] [4] [5] [6] [7] [8] [9]. One of the most prevalent AQM algorithms is RED [1], which detects congestion using an exponentially weighted moving average of the queue length, Õ, and drops or marks packets proportional to Õ at a router buffer before the buffer overflows. ...
... Considerable research has been undertaken on active queue management (AQM) for congestion control in TCP networks. The Random Early Detection (RED) gateways and many other algorithms have been proposed for AQM123456789. One of the most prevalent AQM algorithms is RED [1], which detects congestion using an exponentially weighted moving average of the queue length, Õ, and drops or marks packets proportional to Õ at a router buffer before the buffer overflows. ...
The virtual rate control (VRC) algorithm has been proposed for active queue management (AQM) in TCP networks. This algorithm uses an adaptive rate control instead of queue length control in order to respond quickly to traffic change with high utilization and small loss. By introducing the notion of virtual target rate, the VRC algorithm can maintain an input rate around the target rate, while attempting to regulate the queue length. In this paper, we analyze the stability of the VRC algorithm in a linearized model. From the results of our analysis, we provide a design guideline for the system to remain stable. We show the validity of our analysis and the effectiveness of the VRC algorithm compared to RED, PI, REM and AVQ algorithms through ns-2 simulations.
The stabilizing random early detection (RED) congestion control algorithm in transmission control protocol (TCP)-IP networks is a control theory problem. Significant attention has been drawn to this problem in the networking and control theory research communities. In this paper, we use a nonlinear dynamic model of the TCP RED congestion control algorithm to analyze and design active queue management (AQM) control systems. A linearized model of RED behavior around its nominal operating point which implicitly includes the delay in the control signal is derived. It is assumed that the system model is corrupted at the input and output by zero mean white Gaussian noise signals. An optimal state feedback stochastic controller is designed for the linearized model of the system in conjunction with a Kalman filter for state estimation. To illustrate the proposed design methodology, simulations results are presented and discussed. The proposed stochastic controller is applied to the nonlinear model of the system; Simulation results indicate that the proposed controller keeps the queue length bounded in an appropriate stochastic sense. Copyright © 2009 John Wiley & Sons, Ltd.
Learn to Design High Performance Switches and Routers for Today's Ever Growing Internet Traffic As Internet traffic continues to grow, and demands for quality of service become more stringent, researchers and engineers can turn to High Performance Switches and Routers for tested and proven solutions. This text presents the latest developments in high performance switches and routers, coupled with step-by-step design guidance. More than 550 figures and examples enable readers to grasp all the theories and algorithms used for design and implementation. The authors begin with an examination of the architecture of the Internet, as it is now and as it will be in the future. Then, they examine router architectures and their building blocks, and the challenging issues involved in designing high performance, high-speed routers. Examples of commercial high-end routers are provided. Next, the authors discuss the main functions of the line cards of a core router, including route lookup, packet classification, and traffic management for quality of service control. The bulk of the text is then dedicated to packet switching designs. Coverage includes the various available architectures, algorithms, and technologies. Among the topics covered, readers will find detailed discussions of the latest innovations in electrical and optical packet switching. The final chapter discusses state-of-the-science commercial chipsets used to build routers. Readers learn their architecture and functions, using the theories and conceptual designs presented in the previous chapters as a foundation. Although implementation techniques for switches and routers will continue to evolve, the fundamental theories and principles of this text will serve readers for years to come. In addition to bringing researchers and engineers up to date with the latest designs, this text, with its focus on illustrations and examples, is an ideal graduate-level textbook.
This paper describes the Dynamic-RED (DRED) algorithm, an active queue management algorithm for TCP/IP networks. In random early detection (RED), one of the goals is to stabilize the queue lengths in routers. However, the current version of RED does not succeed in this goal because the equilibrium queue length strongly depends on the number of active TCP connections. Using a simple control-theoretic approach, DRED randomly discards packets with a load-dependent probability when a buffer in a router gets congested. Over a wide range of load levels, DRED is able to stabilize a router queue occupancy at a level independent of the number of active TCP connections. The algorithm achieves this without estimating the number of active TCP connections or flows and without collecting or analyzing state information on individual flows. The benefits of stabilized queues in a network are high resources utilization, bounded delays, more certain buffer provisioning, and traffic-load-independent network performance in terms of traffic intensity and number of connections.
In this paper we study a previously developed linearized model of
TCP and active queue management (AQM). We use classical control system
techniques to develop controllers well suited for the application. The
controllers are shown to have better theoretical properties than the
well known RED controller. We present guidelines for designing stable
controllers subject to network parameters like load level propagation
delay etc. We also present simple implementation techniques which
require a minimal change to RED implementations. The performance of the
controllers are verified and compared with RED using ns simulations. The
second of our designs, the proportional integral (PI) controller is
shown to outperform RED significantly
The ability of proportional integral (PI) and proportional integral derivative (PID) controllers to compensate many practical industrial processes has led to their wide acceptance in industrial applications. The requirement to choose either two or three controller parameters is perhaps most easily done using tuning rules. A summary of tuning rules for the PI control of single input, single output (SISO) processes with time delay is provided in this paper.
This paper presents Random Early Detection (RED) gateways for congestion avoidance in packet-switched networks. The gateway detects incipient congestion by computing the average queue size. The gateway could notify connections of congestion either by dropping packets arriving at the gateway or by setting a bit in packet headers. When the average queue size exceeds a preset threshold, the gateway drops or marks each arriving packet with a certain probability, where the exact probability is a function of the average queue size.
RED gateways keep the average queue size low while allowing occasional bursts of packets in the queue. During congestion, the probability that the gateway notifies a particular connection to reduce its window is roughly proportional to that connection's share of the bandwidth through the gateway. RED gateways are designed to accompany a transport-layer congestion control protocol such as TCP. The RED gateway has no bias against bursty traffic and avoids the global synchronization of many connections decreasing their window at the same time. Simulations of a TCP/IP network are used to illustrate the performance of RED gateways.
The ability of proportional integral (PI) and proportional integral derivative (PID) controllers to compensate many practical industrial processes has led to their wide acceptance in industrial applications. The requirement to choose either two or three controller parameters is perhaps most easily done using tuning rules. A summary of tuning rules for the PID control of single input, single output (SISO) processes with time delay is provided in this paper.
To help improve the performance of congestion avoidance protocols like Transmission Control Protocol (TCP) and to limit the impact of non-adaptive UDP-based applications, the Internet Engineering Task Force has recommended the widespread deployment of active queue management schemes. We describe in this paper a new Random Early Detection (RED) algorithm for congestion control called Dynamic-RED (DRED) from a gradient optimization perspective. One of the goals of RED schemes is to stabilize the queue lengths in routers. However, the current version of RED does not succeed in this goal because the equilibrium queue length strongly depends on the number of active TCP connections. Using a simple optimization technique, DRED randomly discards packets with a load-dependent probability when a buffer in a router gets congested. DRED is also able to stabilize a router queue occupancy at a level independent of the number of active connections over a wide range of load levels. This is done without estimating the number of active TCP connections or flows and without collecting or analyzing state information on individual flows.