大带宽时延积网络的拥塞控制算法研究

发布时间：2018-04-22 21:25

本文选题：大带宽时延积 + 拥塞控制　；参考：《北京邮电大学》2014年硕士论文

【摘要】：随着主干网带宽的大幅度提升、远程网络连接的应用以及卫星网络的普及等,这些网络应用均具有大带宽时延积特征,已成为未来网络的发展趋势。带宽时延积,即带宽与时延的乘积,是一个能够反映网络传输速率的重要指标。在网络高速发展的背景下,原有的拥塞控制算法面临着新的挑战,因此设计适用于大带宽时延积网络的拥塞控制算法具有十分重要的理论意义。针对这种不足,本文从以下两个方面展开论文的研究工作：首先,从互联网使用最为广泛的AIMD算法出发,对其中基于时延的Vegas算法进行了研究和改进。论文分别从慢启动和拥塞避免两个阶段分析了Vegas算法在大带宽时延积网络下存在的问题。在慢启动阶段,本文分析了产生慢启动结束过快的原因,引入了速率控制因子作为调制窗口增长速率的因素,从而有效解决了慢启动结束过快的问题；在拥塞避免阶段,通过动态地调整窗口的变化速率,避免了长时间窗口处于线性增减的问题。本文将该算法命名为Vegas-f,并通过NS2仿真验证了算法的性能,仿真结果表明算法拥有更短的收敛时间,更好的协议间公平性。其次,从MIMD算法的出发,文章分析了目前主要针对大带宽时延积网络的一种拥塞控制协议FAST TCP的性能及存在的问题,提出了改进方案。FAST TCP是一种比较新的拥塞控制协议,本文首先分析了其对网络的依赖性以及公平性方面的不足,其中公平性的不足是由于不同时间段的数据流造成的。通过分析,不同时间段的数据流会造成最小链路时延估计不准确,由此,文章提出了一种由窗口以及RTT变化来调节链路最小时延的改进算法,并且通过与传统算法进行仿真验证对比,从侧面证明了改进算法具有更好的协议内公平性。以上从AIMD和MIMD角度改进的拥塞控制算法,为大带宽时延积网络的应用发展提供了有效的理论支持。
[Abstract]:With the rapid increase of backbone network bandwidth, the application of remote network connection and the popularization of satellite network, these network applications have the characteristics of large bandwidth delay product, which has become the trend of future network development. Bandwidth delay product, which is the product of bandwidth and delay, is an important index which can reflect the network transmission rate. With the rapid development of the network, the existing congestion control algorithms are facing new challenges, so it is of great theoretical significance to design congestion control algorithms suitable for large bandwidth delay product networks. In view of this deficiency, this paper starts the research work from the following two aspects: Firstly, based on the most widely used AIMD algorithm on the Internet, the delay-based Vegas algorithm is studied and improved. In this paper, the problems of Vegas algorithm in large bandwidth delay product networks are analyzed in terms of slow start and congestion avoidance. In the stage of slow start, this paper analyzes the causes of slow start end too fast, introduces the rate control factor as the factor of modulation window growth rate, thus effectively solves the problem of slow start end too fast. By adjusting the changing rate of window dynamically, the problem of long time window in linear increasing or decreasing is avoided. The algorithm is named Vegas-f. and the performance of the algorithm is verified by NS2 simulation. The simulation results show that the algorithm has shorter convergence time and better inter-protocol fairness. Secondly, starting from the MIMD algorithm, this paper analyzes the performance and existing problems of a congestion control protocol FAST TCP, which is mainly used in large bandwidth delay product networks, and proposes an improved scheme. Fast TCP is a new congestion control protocol. This paper first analyzes its dependence on the network and the lack of fairness, in which the lack of fairness is caused by the data flow in different time periods. Through analysis, the data flow in different time periods will cause the minimum link delay estimation to be inaccurate. Therefore, an improved algorithm is proposed to adjust the minimum link delay by changing the window and RTT. Compared with the traditional algorithm, the improved algorithm has better intra-protocol fairness. The above improved congestion control algorithms from the point of view of AIMD and MIMD provide an effective theoretical support for the application and development of large bandwidth delay product networks.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】