论文标题:Internet拥塞控制相关算法研究及仿真分析
Research on and Simulation Analysis of Internet Congestion Control Related Algorithms
论文作者 刘昭辉
论文导师 陶宏才,论文学位 硕士,论文专业 计算机应用技术
论文单位 西南交通大学,点击次数 141,论文页数 85页File Size2518k
2005-06-01论文网 http://www.lw23.com/lunwen_26080952/ 拥塞控制;TCP友好拥塞控制;AQM;CSFQ;仿真分析
Congestion control; TCP-Friendly congestion control; AQM; CSFQ; Simulation analysis
随着新型网络应用的不断涌现和用户数量的迅速增长,Internet的流量正急剧增长,越来越严重的拥塞问题暴露出来。Internet中的拥塞控制对保证Internet的稳定性具有十分重要的作用,使用拥塞控制是其它QoS机制正常工作的必要前提。 TCP是Internet上最主要的传输协议,当前Internet的稳定性主要归功于它所采用的端到端拥塞控制。尽管TCP非常适合于诸如批量数据传输应用,但它不适合于实时应用。为了支持诸如流媒体的实时应用传输,研究人员提出了许多TCP友好拥塞控制机制。本文对这些拥塞控制机制进行了介绍,同时对其性能进行仿真分析。 基于端点的机制实施容易,但它对拥塞的检测是通过拥塞的效果来实现的,在公平性和应付行为不端的信源方面表现较差。因此,研究人员提出了在拥塞发生的地方即路由器来检测拥塞。本文对两种典型的路由器支持的拥塞控制机制AQM(Active Queue Management)和ECN(Explicit Congestion Notification)进行介绍,并对几种典型的AQM算法进行了仿真分析。 TCP拥塞控制机制存在的另一个问题是:TCP难以实现相互竞争的连接之间的公平带宽共享。目前,实现公平带宽共享的机制主要有三种:依流调度机制、依流丢弃机制和无状态公平队列算法。其中,无状态公平队列算法与前两种机制相比,实现复杂度低,具有较好的扩展性。核心无状态算法是用于实现公平带宽分配的典型算法,它在降低算法实现复杂度的同时保留了较好的公平性,但它仍然存在诸多需要改进之处。本文作者提出了一种结合队列长度的CSFQ(Core-Stateless Fair Queuing)改进算法,并对其性能进行了仿真分析。改进的算法能够达到近似公平带宽分配,在保持CSFQ其它优点的基础上,进一步改善了总体吞吐量,减少了分组转发时延,并更有效地利用了链路带宽,且仍能避免拥塞的产生。尤其对小流量和突发性/间歇性流量,该改进算法在性能上有显著提高。
With the emergence of many new network applications and rapid expansion of the user, Internet traffic has been increasing dramatically, and the network congestion appears to be a more and more critical issue. Congestion control mechanisms are very important in guaranteeing the stability of Internet. Moreover, a suitibale congestion control mechanism becomes the necessary premise for other QoS (Quality of Service) mechanism to work effectively.TCP is the dominant transport protocol on Internet, and the current stability of Internet depends on its end-to-end congestion control, which uses an AIMD (Additive Increase Multiplicative Decrease) algorithm. Although TCP congestion control is appropriate for applications such as bulk data transfer, it is not suitable for real-time applications. In order to support the-transfer of real-time application such as streaming multimedia, various TCP-Friendly congestion control mechanisms have been proposed. In this thesis, some of these mechanisms are described, and the performance of them are compared and analyzed through simulation.Congestion control mechanisms based on end point are easy to be implemented, but a problem with them is that the congestion is detected through the effects of congestion rather than the congestion itself. And also there are problems with fairness and non-compaliant sources. Therefore, it seems logical to palce the congestion control mechanism at the location of congestion, i.e, the router. Two typical schemes, that is, AQM (Active Queue Management) and ECN (Explicit Congestion Notification), are introduced in the thesis. The performance of some typical AQM algorithms is also analyzed through simulation.Another problem with TCP"s congestion mechanism is that it is difficult to achieve fair bandwidth sharing among various competing connections. Currently three approchs are used to achieve fair bandwith sharing, that is, per-flow scheduling algorithm, per-flow dropping algorithm and stateless fair
|
