QQ客服 官方微博 反馈留言

题名	分散最优的端到端网络拥塞控制机制研究
作者	刘伟荣
学位类别	工学博士
答辩日期	2007-05-23
授予单位	中国科学院研究生院
授予地点	中国科学院自动化研究所
导师	易建强
关键词	网络拥塞控制 公平性 智能控制 分散最优控制 network congestion control fairness intelligent control decentralized optimal control
其他题名	Research on Decentralized Optimal End-to-End Network Congestion Control
学位专业	控制理论与控制工程
中文摘要	网络拥塞控制机制是维持Internet稳定性和可用性的核心机制之一。随着Internet的发展，其上的网络拥塞机制也需要不断的修改和扩充。近年来已经开展了大量的提高网络拥塞控制机制性能的研究工作，这些工作提出了许多拥塞控制机制的改进方案，包括各种基于窗口的拥塞控制机制和基于流的拥塞控制机制。但由于网络技术的飞速发展和网络拥塞问题的分散特性，在这个领域仍然有许多需要探讨和完成的课题。本论文分析了当前影响网络拥塞控制的关键因素并总结了近几年来的重要研究成果，在此基础之上提出了一系列的改进机制。本论文的主要内容如下： 首先，对网络拥塞的现象、原因和一般性解决思路做了介绍，然后介绍了当前的网络拥塞控制机制即TCP协议并指出了其不足。在这基础上给出了一些主要的改进机制。并阐述了选题背景和论文主要内容。 其次，指出在无线网络上以丢包作为拥塞特征是不合适的。因此扩展了ECN机制使其能提供更多的网络状态信息，也扩展了TCP机制以提供更为平滑的数据流。仿真表明该机制能够在保持较低时延的情况下得到更大的通量。 第三，分析了基于窗口机制的公平性问题，指出基于窗口的拥塞控制协议存在与时延有关的公平性问题，并描述了流控机制中能够较好实现公平性的Kelly流控模型。 第四，针对Kelly模型primal算法线性收敛的特点，提出了一种在链路端引入负价格的网络拥塞流控协议的实现形式，该协议通过对REM机制的扩展使其能传递负的价格。这样的改进可使primal算法变为指数收敛。 第五，针对Kelly模型primal算法稳定时的过载问题，提出了一种在源端使用遗传－模糊控制器修改primal算法和性增加系数的机制，该机制能提高primal算法的适应性和灵活性，使其在瓶颈链路带宽变化时不会产生过载问题。 第六，从网络拥塞控制问题归纳出一类分散最优控制框架，该框架可以分散的形式解决最优控制分配问题，而供需问题和协作问题都可以看作是该问题的具体体现。该框架的提出为解决大规模优化问题提供了一种可行方案。 最后，对取得的研究成果进行了总结，并展望了需要进一步研究的工作。
英文摘要	Network congestion control mechanism is one of the key factors to maintain the stability and practicability of the Internet. The dissertation analyses the key factors of the network congestion problems and summarizes the significant research results. Further, concerning some insufficiency of these results, the dissertation proposes a series of extension to improve them. The main contents of the dissertation are listed as follows. Firstly, the phenomenon, cause, and general solution of the network congestion are introduced. And then, the current main network congestion mechanism, that is, TCP protocol is described. After that, the shortages of the TCP protocol are given and main improvements in recent research are discussed. Then background and main work of this dissertation are described. Secondly, it is pointed that taking packets losses as congestion signal is not proper in wireless network, so it is necessary to use explicit congestion notification to tell the sender the congestion occurring. The conventional ECN code is expanded to provide more state information about the network, and the AIMD mechanism of the TCP protocol is expanded to provide more smooth data flow. The simulation displays the proposed mechanisms can obtain more throughput and can keep low delay time. Thirdly, the fairness of the window-based congestion control mechanisms is analyzed and the unfairness problem depended on delay time of window-based mechanisms is pointed out. To obtain fine fairness, the Kelly fluid-based model is introduced which can realize global fairness. Fourthly, concerning the linear convergence speed of the primal algorithm in Kelly fairness model, a new fluid-based congestion control mechanism is proposed which can allow negative price in the link price updating. The introduced negative price can let the primal algorithm converge with exponential speed. Fifthly, to solve the overload problem of the primal algorithm in Kelly fairness model when the network is in steady state, a Genetic Algorithm-based Fuzzy Controller (GA-Fuzzy controller) is proposed which is used to adjust the additive increase factor of the primal algorithm in sender. The GA-Fuzzy controller can improve the adaptability and flexibility of the primal algorithm, so the aggregative data flow in bottleneck link does not exceed the link capacity even though the bottleneck link capacity changes. Sixthly, a decentralized optimal control framework is given based on network congestion control problem. The framework can solve the optimal control allocation problem by decentralized method. Many applications such as product-consume problem or coordination problem can be taken as one of the optimal control allocation problems. Under the framework, the design of the decentralized control scheme is given. The framework can be extended to solve general large-scale optimal problem with decentralized method. Finally, the obtained results are summarized and future work is addressed.
语种	中文
其他标识符	200418014628091
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/5968]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	刘伟荣. 分散最优的端到端网络拥塞控制机制研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2007.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。