【作者】 刘拥民；

【导师】 蒋新华；

【作者基本信息】 中南大学 ， 控制科学与工程， 2010， 博士

【摘要】 随着Internet的飞速发展,用户数量的急剧增加,新的网络应用不断涌现,虽然网络带宽以摩尔定理的速率增长,但仍无法满足人们对带宽资源的需求。拥塞控制问题在新的网络运营环境下继续成为研究热点。同时,针对层出不穷的不良行为甚至恶意行为流造成的网络拥塞,TCP拥塞控制机制显得力不从心,迫切需要研究新的拥塞控制理论与方法。拥塞控制的目标就是高效、公平地利用网络资源,提高网络的综合性能和服务质量。所以有效地解决拥塞问题是改善网络系统性能、提高网络通讯服务质量的主要手段；并且网络拥塞控制也是当前计算机网络和控制理论交叉领域研究的一个热点课题,不仅具有重要的理论研究背景和意义,同时具有广泛的应用价值。本文主要工作如下：1、在TCP拥塞控制理论上采用流体模型,建立了新的Reno拥塞控制算法延迟微分方程的动态模型,并对该Reno算法模型控制器的局部鲁棒性进行了分析,结果发现Reno协议的一个自身鲁棒性缺陷：在网络延迟猛增或者是链路带宽很大时Reno将变得不稳定。这使得Reno不适合下一代Internet工作环境,然后通过基于包模型的网络仿真实验平台验证了理论推导结果。2、基于TCP源端算法在稳定状态下的循环模型,详细分析了Vegas在公平性能方面的不足,设计了F-Vegas(Fairness-Vegas)算法。所设计的算法可以有效的解决Vegas连接在网络路由改变造成的吞吐量持续下降问题,同时有效的改善Vegas与Reno之间竞争的公平性,通过仿真验证了F-Vegas的高效性和公平性。3、针对Vegas在非对称链路上出现反向通路拥塞导致的TCP连接吞吐量劣化问题,提出了改进算法E-Vegas。E-Vegas利用新的时延测量方法来估计前向通路的可用带宽,有效的提高了传统Vegas连接的吞吐量,同时也有效降低了算法执行的复杂度。4、从协议工作性能的角度对TFRC协议进行了分析,得出了TFRC不适合MANET网络环境的结论。针对此问题提出了Vegas Virtual TFRC协议。与TFRC协议不同,Vegas Virtual TFRC采用Vegas隐式检测策略判断拥塞,并通过虚丢包指示(VLPN)报告拥塞。这两种技术很好的克服了TFRC自身的设计缺陷,很大程度上屏蔽了非拥塞丢包对连接吞吐率的影响,改善了TFRC在MANET网络中的数据传输性能。更多还原

【Abstract】 With the rapid development of Internet size and the explosive increase of the number of users, the new various network applications appear almost every day. Although the network bandwidth increases under Moore rule, it never meets the bandwidth resources requirement of people. Congestion control problem in the new networks environment continues to be the hotspot in the researcher circles. At the same time, more and more users and various applications lead to that traffic increases doubly in the Internet, which makes network congestion frequent. Moreover multiform bad even malice flux behaviors make it more serious. It is difficult for TCP congestion control mechanism to meet the network requirement. So it is imperative to research new congestion control theory and method. The goal of congestion control is to make full use of the network resources effectively and fairly, and to improve the network integrated performance and quality of service. So, the network congestion control is the main way to improve the network performance and reform the quality of service. And the research about network congestion control problem has also been a current hot subject in both computer network and control theory, which makes great sense in theory as well as a wide range of practical application.The main research works and results are listed as follows:1、To the problem of analytic model of the TCP congestion control protocol, using the fluid-flow approximation which we simulate the dynamic behavior of the Reno congestion controller. By means of the theory of function differential equation, we analyze the local Robustness of this model and derive a sufficient Robustness condition. The result indicates that Reno algorithm in the Internet today becomes unstable when delay increases, or when link capacity increases. The instability of Reno makes it unsuitable for high-speed large capacity of Next Generation Internet (NGI), which is inherent. The simulation results coincide with the theoretical results, which demonstrated by the well-known ns-2 simulator.2、Based on the cyclical model for the steady state behavior of the source algorithm, the limits of Vegas are analysis in detail. A new adaptive mechanism for Vegas called the F-Vegas (Fairness-Vegas) algorithm was designed, which can solve the continuous decline in throughput by rerouting. Simulation show that the F-Vegas is able to obtain a fairer share of the network bandwidth when competing with other TCP flows such as Reno. The performance and fairness of the designed algorithm are verified using NS simulations.3、E-Vegas is presented to solve the unnecessary throughput deterioration of Vegas when congestion occurs in the backward path. E-Vegas estimates the available bandwidth of the forward path by a new forward path measurement. The E-Vegas can effectively improve the connection throughput and greatly reduces the complexity of the implementation of the algorithm.4、From the perspective of protocol design, the paper reach the conclusions that TFRC is unsuitable for the MANET environment. According to the problem of TFRC, an enhancement for TFRC called Vegas Virtual TFRC was present. Unlike TFRC, it estimates the congestion by implicit Vegas detection strategy and notifies the congestion by Virtual Loss Packet Notification (VLPN). The two techniques preferably overcome the defects of TFRC, shield the effect of non-congestion packet loss on the connection throughput to a large extent, and improve the performance of TFRC in the MANET.更多还原