Delfin Y. Montuno

A control theoretic approach to active queue management

Abstract 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.

An optimization-oriented view of random early detection

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.

Design and stability analysis of a rate control algorithm using the Routh-Hurwitz stability criterion

Feedback delays arise in the control of a computer network-from the information transfer process itself and from the processing of control signals at the network nodes. Flow control of data sources in a computer network often results in a time-delayed control problem. Feedback delay reduces the stability of a system. In this paper we discuss how to use the Routh-Hurwitz stability criterion to design and analyze the stability of a flow control algorithm with feedback delay.

Enhancing TCP performance with a load-adaptive RED mechanism

This paper describes a technique for enhancing the effectiveness of RED by dynamically changing the threshold settings as the number of connections land system loadr changes. Using this technique, routers and switches can effectively control packet losses and TCP timeouts while maintaining high link utilization. Copyright

Rate-based proportional-integral control scheme for active queue management

Most high-speed links do not have adequate buffering and as a result Active Queue Management (AQM) schemes that utilize queue size information for congestion control cannot be effectively applied on these links. A high-speed link will, typically, have small buffers in relation to the bandwidth-delay product of the link. In this paper we argue that rate-based AQM schemes be used for such links. The goal here is to match the aggregate rate of the active TCP connections to the available capacity while maintaining minimal queue size and high link utilization. The AQM scheme described here employs a Proportional-Integral (PI) control strategy and explicitly takes into account the time delay in the control process.

Enhancing network performance with TCP rate control

In TCP (transmission control protocol), congestion control as well as error recovery are implemented by a sliding window. The dynamics of TCP (specifically, a mismatch between the TCP window and the bandwidth-delay product of the network) can sometimes cause the network switches or routers to accumulate large queues, resulting in buffer overflows, reduced throughput, unfairness and underutilization. It is generally accepted that there is a limit as to how much control can be accomplished from the congestion control mechanisms in the end systems. Some mechanisms are thus needed in the intermediate network elements to complement the endpoint congestion avoidance mechanisms. Network layer enhancements such as scheduling mechanisms and packet drop policies have been proposed which are aimed at improving fairness and throughput of the competing endpoint applications. We describe a new TCP rate control scheme based on a simple recursive algorithm. The idea behind the algorithm is to match the network load to the available resources by modifying at an intermediate network element, the receivers advertised window in TCP acknowledgments returning to the sources. The scheme can be implemented in a router or switch for bandwidth management and does not require knowledge of network delays or maintenance of the per-flow state.

Clock recovery based on packet inter-arrival time averaging

Accurate source clock recovery is an important element of circuit emulation services (CES) over packet networks. A well-known and widely implemented technique for clock recovery in CES is the one that is based on packet inter-arrival time (sometimes called time difference of arrival (TDOA)) averaging. The technique is very simple to implement but provides good performance only when packet losses and packet delay variation (PDV) are very low and well controlled. This technique has not been fully characterized analytically in the literature. In this paper, we provide a full analytical examination of this well-known clock recovery technique. We analyze the effects of correlation of the PDV experienced by the constant bit rate (CBR) traffic stream on the quality of the clock recovered by a receiver. We prove analytically that, for a general input process, high correlation of the PDV produces a large variance of the recovered clock. The paper also describes simple all-digital implementations of the clock recovery scheme using standard digitally controlled oscillators (DCOs).

Effects of control loop delay on the stability of a rate control algorithm

This paper presents exact stability analysis of a rate control algorithm described in Perform. Eval. 2001; 43(2–3):63–94; Int. J. Commun. Systems 2001; 14(6):593–618. The stability regions of the rate control process in the presence of control loop delay are analysed. The rate control process is represented by delay-difference equation and the criteria for asymptotic stability are derived in terms of the control parameters and control loop delay. The analysis shows that the approximate upper bound of the control gain derived in Aweya et al. is very close to the exact bound developed here. Using theoretical calculations performed in the discrete-time domain, we show that as the feedback time delay d increases, the intensity of control (i.e. the control gain α) must decrease in order for the system to remain stable. Copyright

DRED: a random early detection algorithm for TCP/IP networks

It is now widely accepted that a RED [2] controlled queue certainly performs better than a drop-tail queue. But an inherent weakness of RED is that its equilibrium queue length cannot be maintained at a preset value independent of the number of TCP active connections. In addition, REDs optimal parameter setting is largely correlated with the number of connections, the round-trip time, the buffer space, etc. In light of these observations, we propose DRED, a novel algorithm which uses the basic ideas of feedback control to randomly discard 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 benefits of stabilized queues in a network are high resources utilization, predictable maximum delays, more certain buffer provisioning, and traffic-load-independent network performance in terms of traffic intensity and number of connections. Copyright

A simple, scalable and provably stable explicit rate computation scheme for flow control in communication networks

This paper describes fast rate computation (FASTRAC), an explicit rate flow control algorithm for available bit rate (ABR) traffic. Using digital control theory, we develop a simple rate controller for the ABR flow control process. We prove that the controller is stable, fair to all participating sources and configurable with respect to responsiveness. The analysis presented shows that stability of the flow control process depends primarily on two factors, the control update rate and the feedback delay. The implementation of the proposed algorithm is much simpler than other fair rate allocation algorithms. The proposed algorithm demonstrates the ability to scale with speed, distance, different feedback delays, number of users, and number of nodes while remaining robust, efficient, and fair under stressing and dynamic traffic conditions. Copyright