【作者】 秦凯运；

【导师】 谢剑英；

【作者基本信息】 上海交通大学 ， 控制理论与控制工程， 2007， 博士

【摘要】 随着Internet的快速发展,其鲁棒性越来越依赖于Internet所采用的拥塞控制机制。拥塞控制对网络来讲,不仅是个核心问题,也是个非常困难的问题。但随着应用需求的日益丰富和技术的不断发展,研究者开始认识到要想完全依赖实现在终端系统上的策略与算法是很难满足诸如QoS这样复杂的应用需求的。于是,人们开始将部分研究注意力转向网络中的路由器等中间节点设备,期望通过增强它们的功能来实现主机端无法达到的技术目标;为此必须将IP层也纳入到拥塞控制机制中来,通过在路由器中采用一定数据包丢弃或标记策略来提高网络的服务质量,即主动队列管理技术。就拥塞控制而言,网络中间节点可以更加及时,准确地了解网络的拥塞状态,并依此实施有效的资源管理策略,使网络能有效地避免拥塞,或尽早地从严重的拥塞状态中恢复过来。论文旨在设计有助于提高Internet服务质量,增强网络鲁棒性的拥塞控制和队列管理算法。论文首先对目前Internet中普遍采用的TCP拥塞控制算法的性能进行了分析和研究,然后以主动队列管理算法发展过程为线索,详细分析了各个阶段具有代表性的队列管理算法,以及其存在的不足和改进算法;随后针对主动队列管理算法RED存在的参数设置问题,从反馈控制理论的角度对RED算法进行了分析,对于如何使系统在稳定性与动态响应速度两个方面取得良好的平衡给出了具体配置方法,而且对于RED算法的各种参数组合,提出了如何判断其稳定裕度的计算方法。考虑到网络状态参数的时变性,论文根据变结构控制对系统摄动及干扰可呈现出完全鲁棒性的特点,设计了一种基于变结构控制的队列管理算法。随后,针对由于时延造成的稳态误差问题,提出了一种校正方法;此外,根据模糊控制对模型不确定性的适应能力,设计了一种基于模糊逻辑的自校正队列管理算法,随后对该算法做了相应的简化,设计了一种更为实用的自校正队列管理算法,ST算法,并在此基础上将ST队列管理算法进行扩展,设计了一种全新的适于区分服务方式的队列管理算法。论文的主要研究工作和创新包括:根据TCP和式增加积式减小的窗口变化机制,采用效用函数的方法对TCP拥塞控制的公平性进行了分析;针对主动队列管理算法RED存在的参数难以设置问题,对TCP/AQM的非线性模型进行了线性化处理,从反馈控制理论的角度对RED算法进行了分析,对于如何使系统在稳定性与动态响应速度两个方面取得良好的平衡给出了具体配置方法;并对于RED算法的各种参数组合,提出了如何精确判断其稳定裕度的判定方法;针对TCP拥塞控制的数学模型,根据变结构控制对系统摄动及干扰可呈现出的完全鲁棒性,设计了一种基于变结构控制的队列管理算法;针对由于时延造成的稳态误差问题,提出了一种校正方法;将模糊控制与队列管理技术相结合,设计了一种基于模糊逻辑的自校正队列管理算法,该算法不但不需要复杂的参数调节,而且性能优于其它队列管理算法;将模糊自校正队列管理算法进行简化,设计了一种更为实用的自校正队列管理算法,并将该算法进行扩展,设计了一种全新的适于区分服务方式的队列管理算法;更多还原

【Abstract】 As the Internet explodes in size and in the number of users, the robustness of the Internet is highly depended on the congestion control mechanism. To network, congestion control is not only the core issue, but also a very difficult problem. As the development of technology and application demand, researchers began to realize that it is very difficult to meet the demand of complex application, such as QoS, only depending on the policy and algorithm on end system. So some research began to turn to the internal node device, such as routers in the network. By improving their function, they want to reach some technic target which can’t be realized on the end system. Therefore, congestion control mechanism must involve the IP layers. The QoS of network can be improved by adopting some policy of droping or marking packages, which is active queue management technology. In term of congestion control, the internal node could grasp the accurate congestion state in time, and implement the effective resourse management policy, which can avoid congestion or resume from congestion immediately. The purpose of the thesis is to design congestion control and queue management algorithm which is helpful to improve the QoS of Internet and robustness of network.At first, we analysed and researched the performance of TCP, which is the most popular congestion control algorithm. Based on the development sequence of active queue management, we analysed the representative AQM of each stage, including their shortage and improved algorithm. Then, in order to solve the problem of parameter configured, from the point of feedback control theory, we analysed tht RED algorithm, and presented a method of configuring RED parameter, which can obtain a good tradeoff between the stability and dynamic characteristics of system. To the various parameter combinations of RED algorithm, we presented a method to estimate the stability margins accurately. Considering time variety of the state parameter of network, based on the variable structure control, we developed a VS active queue management algorithm. Then, to the probleme of steady-state error, we presented the method to revise the Steady-state error. According to the mathematical model of TCP congestion control, we developed a self-tuning fuzzy queue management algorithm with the consideration of fuzzy controller being well suited for model uncertainty. Through properly simplification of the fuzzy algorithm, we developed a more practical queue management algorithm, i.e. the Self-tuning algorithm. By properly modification, ST can not only be used in networks with in and out packets, but also suitable for network service with different priority class. Specifically, the main work and contributions in this thesis are as follows.Based on the mechanism of Additive Increase Multiplicative Decrease of congestion window, we analysed the fairness of TCP congestion control using the method of utility function.In order to solve the problem of parameter configured, from the point of feedback control theory, we analysed tht RED algorithm, and presented a method of configuring RED parameter, which can obtain a good tradeoff between the stability and dynamic characteristics of system. To the various parameter combinations of RED algorithm, we presented a method to estimate the stability margins accurately.According to the mathematical model of TCP congestion control, we developed a VS active queue management algorithm based on the variable structure control. Then, to the probleme of steady-state error, we presented the method to revise the Steady-state error.We combined the Fuzzy control and queue management technology, and developed a self-tuning fuzzy queue management algorithm with the consideration of fuzzy controller being well suited for model uncertainty. The main advantage of this algorithm is that, the queue length can keep stable in a variety of network environments without the difficulty of parameter configuration.Through properly simplification of the fuzzy algorithm, we developed a more practical queue management algorithm, i.e. the Self-tuning algorithm. With the analysis of RIO, we explore the use of ST in Assured Service architecture. By properly modification, ST can not only be used in networks with in and out packets, but also suitable for network service with different priority class.更多还原