【作者】 杨燕；

【导师】 谭连生；

【作者基本信息】 华中师范大学 ， 计算机应用技术， 2005， 硕士

【摘要】 进入九十年代以来,以IP为基础的Internet呈爆炸式增长,新型网络应用不断涌现,用户数量迅速增加,使Internet的流量急剧增加,由此而引发的网络拥塞已经成为制约网络发展和应用的瓶颈问题。拥塞容易造成传输延迟和吞吐量等QoS(Quality of Service)性能指标下降,严重影响了带宽、缓存等网络资源的利用率,因此有效地解决拥塞问题对于提高网络性能具有重要意义,如何更好地预防和控制拥塞一直是近年来国际上网络研究领域的热点问题。 目前Internet仅提供单一的“best effort”服务,其资源分配主要在用户端进行,利用传输控制协议(TCP)进行端到端的拥塞控制,可以说,Internet的成功在很大程度上依赖于TCP拥塞控制机制的有效执行,但是由于Internet的发展及TCP拥塞控制算法本身存在的问题,使得现有的拥塞控制策略在很多方面已经不能满足人们的需求。研究表明,TCP拥塞控制实质上是一种较保守的策略,它并非在所有的网络条件下都能保证其良好的性能。 Internet中传统的路由器通常采用先来先服务的调度算法以及“弃尾”缓冲管理方法,在指示和控制拥塞方面不提供任何显式的支持。由于在路由器中引入相应的拥塞控制机制,使网络本身参与资源的控制工作可以更有效地实现对拥塞的监测和预防,因此近几年IP拥塞控制策略成为当前网络研究的一个热点,IETF(Internet工程任务组)建议在Internet路由器上采用主动队列管理机制作为IP层参与拥塞控制的手段,随机提前检测(RED)算法是IETF推荐的一种主动队列管理算法。 本文首先概要介绍了Internet中的拥塞现象产生的背景、原因以及实施拥塞控制的必要性,探讨了拥塞控制策略的研究与发展现状,从不同角度对拥塞控制机制进行了分类,并分析了当前拥塞控制中存在的问题。⑧器款氯 第二章着重研究了Internet中TCP的基于窗口的端到端拥塞控制方法以及IP层采用的拥塞控制机制。首先对TCP基于滑动窗口的拥塞控制机制的四个核心算法:慢启动、拥塞避免、快速重传和快速恢复进行了详细阐述,讨论了当前的各种TcP拥塞控制改进方案,如NewReno、SACK、TCP一Vegas等。然后对Intemet路由器采用的“弃尾”算法及其存在的问题进行了分析,介绍了主动队列管理机制的思想,讨论了当前业界在IP拥塞控制方面所做的相关工作,并重点分析了RED及其相关算法。 在保证网络稳定和不发生拥塞的基础上,公平性是算法设计的另一个重要的性能指标。因此,在第三章我们根据网络层显示拥塞指示技术支持,提出一种公平窗口算法,使该算法在IP网络中得以实现。仿真结果证明,此算法在多瓶颈网络环境下能使TCP流达到较好的公平性。 在第四章中,我们探索了DRED(oynamie readom early deteetion)方案的队列稳定性,并根据控制理论和控制参数优化的仿真实验提出了一种优化的参数选择方法,从理论上给出了系统稳定的参数选择范围,有效地解决了以前对RED参数选择的盲目性、随机性和不确定性的问题。而且根据对所进行的理论分析,通过仿真实验验证所提算法的有效性。 在上一章研究的基础上,我们在这一章进一步对主动队列管理算法PD一controlle:的参数选择进行探讨。尽管最近提出的比例、差分随机早期预测 (PD一RED)机制改进了网络性能,但和其他的RED机制一样,仍然存在一个关键的问题:参数调整的问题。而选择适当的控制参数使动态队列达到稳定对任何一个RED机制来说都是非常重要的。因此,本章利用TCP动态流量模型和现代控制理论对主动队列管理机制稳定性进行了分析,从理论上给出了系统稳定时的参数选择范围。而且通过仿真对所讨论的参数选择方法进行了验证,仿真结果表明:所提出的参数优化方法很好的保证了系统稳定性,改进了主动队列管理机制的网络性能。 在第六章中,我们推导出AVQ方案的显示的稳定性条件,对文献[53」中的方案进行了扩展,提出了一种简单、实用的参数调节方案。而且,我们还通过仿真实验进一步证实了所提方案能更有效地使队列长度趋于稳定,具有更多还原

【Abstract】 Internet has experienced an explosively growth since 1990’s. Widespread use of computer networks, as well as the appearance of varied network applications has brought forth network congestion as a significant problem. Congestion often results in the decline of Quality of service (Qos) such as transmission delay and throughput, while the network resource utilization like bandwidth and buffers are also affected seriously. Therefore, it is important to solve the congestion problem effectively for improving network performance. How to avoid and control congestion is one of the most active fields in the computer networks.Current TCP/IP networks only provide best effort services and their resources allocation is mainly on end hosts. The end hosts use Transmission Control Protocol (TCP) to implement end-to-end congestion control. It can be said that the success of today’s Internet primarily relies on TCP congestion control mechanisms. With the development of Internet and the problems in the implementation of TCP, the current TCP strategies couldn’t meet various demands of users any longer. More and more researches indicate that the TCP congestion control mechanisms, while necessary and powerful, are substantially conservative and are not sufficient to provide good service in all circumstances.The traditional technique for managing router queue in Internet usually adopts first in first out (FIFO) scheduling and "Drop tail" queue management algorithms, which have no provision for the detection of incipient congestion when the queue is full. In fact, it will be more effective for detecting and preventing congestion if the routers perform congestion control strategies, so IP congestion control strategies has been discussed widely in recent years. The IETF (Internet Engineering Task Force) has proposed the solution by deploying active queue management in Internet routers.In this paper, Chapter One gives a general introduction to the network congestion control. First the background and the cause of network congestion areintroduced, and then the necessity of implementing congestion control is analyzed. We also discuss the research and development of congestion control strategies, and classify them from different aspects. Finally existing problems in current congestion control are proposed.In chapter two we focus on TCP end-to-end congestion control strategies and the mechanisms implemented in IP routers. First the four kernel algorithms of TCP congestion control are described, and the improved schemes such as New-Reno, SACK and TCP-Vegas are discussed either. Then we introduce "drop tail" algorithm and the principal idea of active queue management. Finally we discuss the related works in IP congestion control, in which RED and ECN algorithm are emphasized.When network stability and congestion nonoccurrence are guaranteed, fairness is an important performance index of the network. Therefore, a new fair window algorithm is proposed in the third part and is realized in IP network in support of ECN mechanism in network layer. The simulation result demonstrates that remarkable fairness among TCP data flow in the multiple bottleneck network can be achieved with the application of this algorithm.Based on control theory, Chapter four provides guidelines for the selection of the control gain for dynamic-RED to stabilize a congested queue at a target and hence to improve network performance. Simulations demonstrate that indeed satisfactory performance can be achieved if the control gain is selected based on the guidelines.On basis of the study of the above chapter, we further explore the control gain selection of another AQM (active queue management) algorithm-PD controller. Though the recently proposed proportional and differential random early detection (PD-RED) scheme has been used in enhancing the performance of queuing networks, one key issue in applications of this scheme as well as of other RED variations is the so-called parameter-tuning problem. The guideline on choosing control gains to meet the stability condition of queue dynamic更多还原