Adaptive neural queue management for TCP networks

Active Queue Management (AQM) is a proven strategy to efficiently maintain queues and ensure high utilization of Transmission Control Protocol (TCP) network resources. The fundamental mechanism is to manage incoming packet rates at a router to prevent incipient network congestion. In this paper, we present an efficient neural network AQM system as a queue controller. The recurrent neural network has a Multi-layer Perceptron-Infinite Impulse Response (MLP-IIR) structure. Three distinct neural AQMs are trained under different network scenarios involving traffic levels. Selecting one of three neural AQMs is based on posterior probability history of traffic level. In addition, we investigate stochastic modeling of the network dynamics by a Dynamic Bayesian Network (DBN). This model allows implementation of a predictive AQM system in which queue dynamics are predicted and used for error prediction via online DBN estimation. Our AQM method is evaluated through simulation experiments both using an Ordinary Differential Equation (ODE) network model and using OPNET^©. The simulation results demonstrate that our adaptive neural AQM outperforms Random Early Detection (RED) and Proportional-Integral-Derivative (PID) based AQM.     

模糊自适应算法在网络主动队列管理中的研究

为了改善由于RED算法控制的主动队列管理策略(AQM)系统引起网络响应时间过长,许多文献提出了基于P/PI控制器的AQM策略。但是,由于网络具有非线性特点,论文设计了一种模糊自适应PID控制器的主动队列管理系统,根据队列长度和队列长度变化通过模糊控制规则在线调整PID控制器参数,并用Matlab进行了仿真,验证了当网络出现突发分组时,该算法能够较快的使分组稳定在队列阀值附近,同时不产生队列震荡。     

PIPIO：一个新的面向区分服务确保转发的主动队列管理算法

RIO是用于支持区分服务确保转发逐跳行为的主动队列管理算法，该算法是对RED算法的简单扩充。由于RED算法的性能对配置参数敏感，因此基于RED算法的RIO算法必然具有配置参数敏感的特点。PI算法是基于控制论的主动队列管理算法，具有队列长度抖动小的特点。PIP算法是PI算法的改进，比PI具有更快的收敛速度。本文基于PIP算法设计了一个新的主动队列管理算法PIPIO。该算法队列长度抖动小，同时能保护高优先级报文。     

Performance evaluation for DRED discrete-time queueing network analytical model

Due to the rapid development in computer networks, congestion becomes a critical issue. Congestion usually occurs when the connection demands on network resources, i.e. buffer spaces, exceed the available ones. We propose in this paper a new discrete-time queueing network analytical model based on dynamic random early drop (DRED) algorithm to control the congestion in early stages. We apply our analytical model on two-queue nodes queueing network. Furthermore, we compare between the proposed analytical model and three known active queue management (AQM) algorithms, including DRED, random early detection (RED) and adaptive RED, in order to figure out which of them offers better quality of service (QoS). We also experimentally compare the queue nodes of the proposed analytical model and the three AQM methods in terms of different performance measures, including, average queue length, average queueing delay, throughput, packet loss probability, etc., aiming to determine the queue node that offers better performance.     

大时滞网络中的拥塞控制算法

主动队列管理(AQM)通过网络中间节点有目的的分组丢弃实现了较低的排队延时和较高的有效吞吐量,是近年来TCP端到端拥塞控制的一个研究热点.已有的大多数AQM算法在设计过程中都没有充分考虑到大时滞对算法性能的影响.首先通过仿真试验证实了已有的几种典型算法控制的队列在大时滞网络中无一例外地出现了剧烈的振荡,导致瓶颈链路利用率下降和延时抖动加剧.为此,在进行了适当模型拟合处理的基础上,应用控制理论中的内模补偿原理设计了鲁棒的延时补偿主动队列管理(delay compensation-active queue management,简称DC-AQM)算法,克服了大时滞给队列稳定性造成的不利影响.仿真实验结果表明,新算法在大时滞小期望队列长度的网络配置中表现出的综合性能明显优于已有的算法,链路利用率是其他算法的3～4倍.     

一种链路负载自适应的主动队列管理算法

随机早检测(random early detection,简称RED)是IETF推荐部署的主动队列管理(active queue management,简称AQM)算法.RED存在参数难以配置、无法得到与流量无关的平均队长等问题.ARED(adaptive RED)是RED的自适应版本,它根据平均队长动态调节最大标记概率参数,从而得到稳定的平均队长.但ARED没有克服瞬时队列长度振荡问题,且在动态流量环境下性能明显降低.分析了ARED性能问题的原因,并提出了一种链路负载自适应的主动队列管理算法LARED(load adaptiveRED).LARED具有两个特点:自适应链路负载、快速响应队长变化.分析和仿真实验表明,与ARED等其他AQM算法相比,LARED在保持高链路利用率和低时延的同时可以得到稳定的瞬时队长,并且具有良好的响应性和鲁棒性.     

基于区分服务的主动队列管理算法研究

分析了几种主动队列管理算法。RIO算法是用于支持区分服务确保转发逐跳行为的主动队列管理算法,它是对RED算法的简单扩充,但是该算法的性能对配置参数敏感。PI算法是基于控制论的主动队列管理算法,具有队列长度抖动小的特点。PIP算法是PI算法的改进,比PI算法具有更快的收敛速度。为了更好地满足AF PHB的要求,基于PIP算法,结合三色标记器的功能,提出一个新的主动队列管理算法PIPGYR（PIP with Green ＆ Yellow ＆ Red）。通过仿真验证,该算法队列长度抖动小,同时能够保护高优先级分组。     

随机早期检测主动队列管理算法的改进研究

主动队列管理（Active Queue Management）算法是近几年网络研究的重点。为了改进和完善现有的AQM算法和设计更好的新算法,需要对主动队列管理机制的实现方法和性能特性进行深入研究,详细讨论了随机早期检测（Random Early Detection）主动队列管理机制的关键技术问题,研究了近年来对RED算法的主要改进算法,总结了这几种算法的优缺点及其有待改进之处,并应用网络仿真器（Network Simulator）对实际网络模型进行了仿真实验,检验了改进算法对网络性能的影响。     

一种模糊自调整的PD-RED算法

RED算法作为第一代主动队列管理技术,能够有效地控制队列长度。然而RED算法在实现中存在着其队列长度依赖于流量负载的变化,网络性能对参数敏感的问题。论文将模糊控制技术与比例微分控制方法相互结合,采用模糊控制器在线调整比例微分控制器参数的方法实现RED算法控制。仿真结果证明,所提出的模糊自调整的PD-RED算法动态响应快,稳态误差小,能够使队列保持期望队列长度。     

RED队列稳态误差分析

主动队列管理在保证较高吞吐量的同时，通过在交换节点上主动丢弃数据包来控制队列长度，从而实现对端到端的延时和抖动的控制．RED算法是目前应用最为广泛的主动队列管理(AQM)算法．RED算法以平均队列长度作为衡量网络拥塞的指标，其参数设置对算法性能有较大影响．利用现代控制工程理论，将RED算法看做一种单位反馈控制系统，并将期望队列长度作为系统输入，将瞬时队列长度作为输出，对该系统的稳态误差进行了分析．实验结果表明在稳定状态下，RED队列的波动受分组丢弃概率函数的斜率影响．在稳定条件边界附近，系统的稳态误差急剧增加．     

A novel self-tuning feedback controller for active queue management supporting TCP flows

Wireless access points act as bridges between wireless and wired networks. Since the actually available bandwidth in wireless networks is much smaller than that in wired networks, there is a bandwidth disparity in channel capacity which makes the access point a significant network congestion point. The recently proposed active queue management (AQM) is an effective method used in wired network and wired-wireless network routers for congestion control, and to achieve a tradeoff between channel utilization and delay. The de facto standard, the random early detection (RED) AQM scheme, and most of its variants use average queue length as a congestion indicator to trigger packet dropping. In this paper, we propose a Novel autonomous Proportional and Differential RED algorithm, called NPD-RED, as an extension of RED. NPD-RED is based on a self-tuning feedback proportional and differential controller, which not only considers the instantaneous queue length at the current time point, but also takes into consideration the ratio of the current differential error signal to the buffer size. Furthermore, we give theoretical analysis of the system stability and give guidelines for the selection of feedback gains for the TCP/RED system to stabilize the instantaneous queue length at a desirable level. Extensive simulations have been conducted with ns2. The simulation results have demonstrated that the proposed NPD-RED algorithm outperforms the existing AQM schemes in terms of average queue length, average throughput, and stability.     

Design and analysis of an ONOFF variable structure controller for AQM routers supporting TCP flows

Active queue management （AQM） can maintain smaller queuing delay and higher throughput by purposefully dropping packets at intermediate nodes. Most of the existing AQM schemes follow the probability dropping mechanism originating from random early detection （RED）. This paper develops a novel packet dropping mechanism for AQM through designing an ONOFF controller applying the variable structure control theory. Because the binary ONOFF controller can considerably simplify the manipulation on the AQM router, it is helpful for implementing the high performance router. The design principles of ONOFF controller are discussed in detail. The guidelines towards parameter settings are presented. The performance is extensively evaluated and compared with other well-known controllers through simulations and theoretical analysis. The results demonstrate that the ONOFF controller is responsive and robust against external disturbances, and is insensitive to variances of the system parameters. Therefore, it is very suitable for the time- varying network system, and at the same time, it can also keep the instantaneous queue length at a desired level with rather small oscillations, which is conducive to achieving the technical objectives of AQM.     

主动队列管理算法的研究

队列管理机制是实现网络拥塞控制的一项重要技术,以往采用的大多都是被动的队列管理机制,而主动队列的管理是根据网络结点的队列长度的变化进行提前丢包,对网络的拥塞进行预先通知,从而减少和避免网络拥塞,提高服务质量.为了对主动队列管理机制进行研究,对IEFT推荐的RED算法作介绍,与传统的被动管理机制Droptail作比较,并且通过网络仿真器NS2对算法进行模拟与分析,指出算法的优缺点,为进一步研究AQM算法提供依据.     

Analysis and design of controllers for AQM routers supporting TCPflows

In active queue management (AQM), core routers signal transmission control protocol (TCP) sources with the objective of managing queue utilization and delay. It is essentially a feedback control problem. Based on a recently developed dynamic model of TCP congestion-avoidance mode, this paper does three things: 1) it relates key network parameters such as the number of TCP sessions, link capacity and round-trip time to the underlying feedback control problem; 2) it analyzes the present de facto AQM standard: random early detection (RED) and determines that REDs queue-averaging is not beneficial; and 3) it recommends alternative AQM schemes which amount to classical proportional and proportional-integral control. We illustrate our results using ns simulations and demonstrate the practical impact of proportional-integral control on managing queue utilization and delay     

LRED: A Robust and Responsive AQM Algorithm Using Packet Loss Ratio Measurement

Active queue management (AQM) is an effective means to enhance congestion control, and to achieve trade-off between link utilization and delay. The de facto standard, random early detection (RED), and many of its variants employ queue length as a congestion indicator to trigger packet dropping. Despite their simplicity, these approaches often suffer from unstable behaviors in a dynamic network. Adaptive parameter settings, though might solve the problem, remain difficult in such a complex system. Recent proposals based on analytical TCP control and AQM models suggest the use of both queue length and traffic input rate as congestion indicators, which effectively enhances stability. Their response time generally increases however, leading to frequent buffer overflow and emptiness. In this paper, we propose a novel AQM algorithm that achieves fast response time and yet good robustness. The algorithm, called Loss Ratio-based RED (LRED), measures the latest packet loss ratio, and uses it as a complement to queue length for adaptively adjusting the packet drop probability. We develop an analytical model for LRED, which demonstrates that LRED is responsive even if the number of TCP flows and their persisting times vary significantly. It also provides a general guideline for the parameter settings in LRED. The performance of LRED is further examined under various simulated network environments, and compared to existing AQM algorithms. Our simulation results show that, with comparable complexities, LRED achieves shorter response time and higher robustness. More importantly, it trades off the goodput with queue length better than existing algorithms, enabling flexible system configurations     

基于智能预测控制的网络拥塞主动队列管理算法研究

路由队列管理是保证网络性能、避免网络拥塞的重要手段,目前采用的主要队列管理方法为被动式队列管理,同时主动式队列管理已经成为近来的主要研究热点. 随机早侦测(RED)作为最早提出的主动队列管理方法,更获得了普遍的关注. 使用严格的数学模型来描述由端系统和网关组成的系统,并进行队列管理性能分析. 提出一种采用快速广义预测控制的RED控制器( FGPC2RED控制器) ,进行网络拥塞控制的研究. 介绍了系统的结构及系统的辨识, 并通过仿真证明了FGPC算法在路由队列管理中应用的可行性,可以有效控制队列长度,避免路由拥塞及减小往返延迟.     

Stability of TCP/RED systems in AQM routers

By applying the time-delay control theory to a TCP/RED dynamic model, this note establishes some explicit conditions under which the TCP/RED system is stable in terms of the average queue length. Then, the stability region is discussed. Finally, the results are illustrated by using ns2 simulations, which demonstrates that it is able to choose an appropriate control parameter max/sub p/ of RED based on the stability conditions derived in this note, to achieve satisfactory network performance. It is found, by comparison, that this improved performance is better than that of three other typical active queue management (AQM) schemes-the random exponential marking (REM), proportional-integral (PI) controller, and adaptive virtual queue (AVQ) schemes.     

On the use of a full information feedback to stabilize RED

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.     

基于时滞补偿的网络拥塞控制策略*

中间节点上的主动队列管理策略在保证较高吞吐量的基础上能有效控制队列长度和端到端时延,利用频率域模型降阶拟合方式建立了TCP流量控制中主动队列管理系统的等效模型,应用控制理论中的内模补偿原理设计鲁棒的延时补偿主动队列管理控制算法,克服了大时滞给对队列稳定造成的不利影响.仿真结果表明,该补偿方法在长时滞小期望队列综合性能明显优于已有的RED,REM,PI等算法,链路利用率大大提高.     

自适应的PIP主动队列管理机制

近年来AQM的研究者提出了多种主动队列管理机制,包括RED,PI,REM,AVQ,PD,SMVS,PIP等,它们之间的主要区别在于丢弃概率的计算方法不同,其中基于反馈校正的PIP是综合性能更为突出的一种算法,但是遗憾的是其参数不能实现自动配置 .结合单神经元自适应PID控制器,为PIP算法建立了自适应的模型,提出一种参数自适应的PIP算法 .通过NS2仿真实验,验证了该算法能提高链路利用率和降低报文丢失率,有效缓解了根据特定网络条件配置算法参数的问题 .结合PI,REM,AVQ,PD等AQM算法,讨论了该自适应模型在其他AQM机制中的推广 .