基于网络测量TCP协议设计.docVIP

基于网络测量的TCP协议设计 专业名称：计算机科学与技术 摘 要 随着信息化浪潮和互联网规模的扩大，网络拥塞已经成为了一个令人无法忽视的问题。由于当今网络的基础架构是TCP/IP协议族，所以学界关注的焦点在于TCP拥塞控制。其控制机制是根据反映网络负载的隐含信号来动态地调整滑动窗口。原理是发送方即源端把丢包事件作为拥塞信号，当检测到拥塞发生后就相应地减少发送窗口大小从而达到缓解和控制拥塞的目的。然而，由于网络服务类型的转变，传统TCP拥塞控制机制受到了挑战。 本文的主要内容是Multi Phase算法，即分段逼近的慢启动算法。首先，文中提到当前使用的TCP版本Tahoe、Reno、Vegas其参数设置灵活性比较小，适应能力比较低，本论文提出采用网络测量技术进行网络可用带宽估计，利用带宽值为拥塞控制提供参考，从而更好地适应网络容量变化的要求；其次，文中提到TCP拥塞控制的慢启动机制中所采用的指数增长算法对于网络流量具有较大的冲击，会引起可用容量迅速枯竭，本论文提出采用分段逼近的慢启动算法，于是降低了突发性的洪峰信息量，避免了拥塞的频繁发生；最后，文中提到的分段逼近的渐进式增长规律在某些特殊场合，比如说Web页面的Java Applet、Flash下载时，该算法的效率不高，其持续时间过长，本论文提出采用设置一个增长因子来调节拥塞窗口的增长幅度，从而动态地适应不同类型的服务需要，提高了网络传输效率，取得了性能与公平的折衷。 本论文中采用基于LINUX平台的NS-2模拟器进行相关算法的实验仿真。从定性的角度来说，本论文提出的算法达到了预期的效果，实现了网络性能的提高，改进了网络传输效率，取得了一定意义上的创新。 关键词 TCP拥塞控制；慢启动阶段；带宽估计；分段逼进算法 ABSTRACT As the information tide spreads wide and the Internet blooms splendidly, the congestion of networks turns into a inevitable problem. Nowadays the basic infrastructure of networks is the suite of TCP/IP , so institutions mainly take a focus on TCP congestion control. The mechanism dynamically adjusts the sliding window according to internal signs for the load of networks. TCP sender which is the source end-point accordingly shrinks the sending window in order to delay and control the congestion when detecting the rise of the congestion which takes packet loss as signals. The traditional TCP congestion control mechanism comes into a challenge with the transformation of service types on networks. The thesis is centered on the Multi Phase algorithm, that is, the phase-divided asymptote algorithm. Firstly, the TCP versions including Tahoe, Reno, Vegas referred to have lack of flexibility on parameters configurations and adaptive capability, so the proposal discussed adopts the bandwidth estimation based on the networks measurement which provides references for congestion control to meet requirements for fluctuation of networks capacity ; then, the exponential increase law in the slow start mechanism referred to takes an impact