【作者】 黎敏；

【导师】 李力；

【作者基本信息】 南昌大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 互联网在过去的十几年中经历了爆炸式的增长,充分证明了其设计上的成功。然而随着互联网用户数量的膨胀,网络的拥塞问题也越来越严重。例如由于队列溢出,互联网路由器会丢弃约10%的数据包。Jacobson自1988年以来提出拥塞概念以来,拥塞控制得到极大关注,各种拥塞控制理论相继提出。从目前来看TCP的AIMD的拥塞控制策略对Internet的正常运行起到了至关重要的作用。针对路由器的拥塞控制也是目前的一个研究热点。 近年来视频点播,远程教育,视频会议等实时流媒体应用大量增多,这些实时流媒体应用对时延要求敏感,不要求可靠传输,通常采用UDP协议进行传输。由于UDP协议本身没有拥塞控制,因此在这些应用中加入拥塞控制是必要的。在流媒体应用中加入拥塞控制一方面可以减少丢包率,使得流媒体应用在拥塞状态下仍能相对平稳的运行从而提高流媒体应用的QoS,另一方面也有助于缓解网络的拥塞。本文讨论了两种常用的针对UDP应用的拥塞驱动方法:基于丢包率和了基于RTT驱动的算法,并指出各自的不足。在此基础上提出一种在接收端通过单向延时预测拥塞的算法(Adaptive One Way Delay Congestion Control,AOWDCC)。并且证明了异步时钟(采用收发端各自的时钟)方式下AOWDCC的正确性。与前两种拥塞驱动机制相比,AOWDCC算法最大限度的缩短了拥塞反馈时间间隔,拥塞判断的实时性大大增强。另外传统TCP拥塞机制的AIMD在拥塞时发送窗口减半的做法不能满足实时媒体流负载平滑的要求,本文提出了基于拥塞级别的线性函数动态调整发送速率的控制机制,极大的平滑了发送速率的调整。最后采用NS模拟器,对基于丢包率的拥塞控制,基于RTT的拥塞控制和AOWDCC算法的性能进行仿真,验证了AOWDCC的优越性。更多还原

【Abstract】 Internet has gone violent growth in the past more than ten years, and the design of Internet has been fully proved success. With the growth of the Internet business , more and more congested happened. The congested control strategy of TCP has played an essential role to the normal running of Internet seen from the present.In recent years a large amount media flow application increased such as video play, remote education, videoconference etc. These applications usually use Udp transport. Udp self has no congestion control mechanism, so these kind application need add congestion control mechanism. One hand introduce congestion control mechanism in it can reduce the lost packet and improve the its Qos, on the other hand it also can contribute alleviate network congestion. The paper discuss two congestion control mechanism base on udp, one is base on ratio of lost packet another is base on RTT, and point out them shortcoming. The paper proposing congestion control mechanism based on One-Way delay. This mechanism reduction addition packets greatly and can quick estimate network congestion, quick adjust send rate and reduce the ratio of lost packet. The congestion control mechanism based on One-Way delay adjust sending rate dynamically according a liner function of congestion level, smoothly adjust sending rate. Finally paper make test in NS (Network Simulate), the result of test proof the superiority of the mechanism base on One-Way delay.更多还原