路由器拥塞控制中RED算法的优化研究

发布时间：2018-04-25 15:27

本文选题：拥塞控制 + RED　；参考：《吉林大学》2014年硕士论文

【摘要】：当今，互联网发展中遇到的主要问题之一，是网络拥塞所导致的资源缺乏和流量分配不均，而拥塞问题还会造成网络数据传输的延迟，进而加剧网络拥塞的程度，更严重的情况下将会导致网络崩溃，上述现象引起了研究者对网络拥塞控制的高度重视。目前，网络拥塞问题仍然是不可避免的，如果仅仅依靠源端的拥塞控制机制，那么会大大降低网络资源的利用率。解决这一问题最有效的办法就是队列管理算法，队列管理算法又分为主动队列管理算法和被动队列管理算法。目前广泛使用的是主动队列管理算法，它通过将路由器中的队列长度控制在一个适当值附近，从而能有效地减少数据包的延时，并且能够提高链路利用率。因此，对主动队列管理算法的研究具有非常重要的意义。随着互联网的飞速发展，现有的TCP拥塞控制机制在很多方面已经无法满足人们的需要，于是人们将拥塞控制研究扩大到网络的中间环节，出现了许多基于路由器的拥塞控制策略。随机早期检测算法RED(Random Early Detection)是网络工程任务组IETF(Internet Engineering Task Force)推荐的基于路由器的拥塞控制算法。本文的主要工作如下：首先，本文详细介绍了RED算法的参数设置，对TCP最大窗口问题进行了分析，并从Misra的随机微分方程模型理论出发，利用数学手段对此模型进行了一系列的处理，最后得到了一种新的估计随机早期检测算法参数——最大丢弃概率的方法。其次，详细介绍了RED算法中重新设置最大丢弃概率的方法：对于一个已经确定的网络，在设置参数时，，最大阈值通常设为最小阈值的2~3倍，期望的队列长度的大小可以根据实际需要设定。如果突发比较强或者对延时比较保守，就将队列长度设得稍小些；如果对丢包比较保守则将队列长度设得偏大些。这样，本文提出的最大丢弃概率估计方法其实只与最小阈值和队列长度有关，当这两个参数确定以后，由推导出的公式即可计算出最大丢弃概率值。最后，通过NS-2(Network Simulation Version2)仿真实验，利用重新设置的最大丢弃概率值对RED路由与DropTail路由的吞吐率和延时进行了比较，并按照相同最大窗口限制和不同最大窗口限制两种条件，对两种算法下的瞬时队列长度、平均队列长度及丢包率进行了比较与分析。最终验证了在重新设置最大丢弃概率后，RED算法瞬时队长、平均队长、丢包率、源端连接数等技术指标均得到优化，从而达到了降低网络拥塞，提高传输质量的目的。
[Abstract]:Nowadays, one of the main problems encountered in the development of the Internet is the lack of resources and the uneven distribution of traffic caused by network congestion, and the congestion problem will also lead to the delay of network data transmission, which will aggravate the degree of network congestion. More serious cases will lead to network collapse, which causes researchers to attach great importance to network congestion control. At present, the network congestion problem is still inevitable. If we only rely on the source congestion control mechanism, it will greatly reduce the utilization of network resources. The most effective way to solve this problem is queue management algorithm, which is divided into active queue management algorithm and passive queue management algorithm. At present, active queue management algorithm is widely used. It can effectively reduce the delay of data packets and improve the link utilization by controlling the queue length in routers near a suitable value. Therefore, the study of active queue management algorithms is of great significance. With the rapid development of the Internet, the existing TCP congestion control mechanism has been unable to meet the needs of people in many aspects, so the study of congestion control has been expanded to the intermediate link of the network. There are many router-based congestion control strategies. Random early Detection (RED(Random Early Detection) is a router-based congestion control algorithm recommended by IETF(Internet Engineering Task Force. The main work of this paper is as follows: First of all, the parameter setting of RED algorithm is introduced in detail, and the TCP maximum window problem is analyzed. Based on the stochastic differential equation model theory of Misra, a series of mathematical methods are used to deal with the model. Finally, a new method to estimate the parameter of the random early detection algorithm, the maximum drop probability, is obtained. Secondly, the method of resetting the maximum drop probability in the RED algorithm is introduced in detail: for a network that has been determined, the maximum threshold is usually set to 2 or 3 times the minimum threshold when the parameters are set. The size of the desired queue length can be set according to actual needs. If the burst is stronger or the delay is conservative, the queue length is set slightly smaller, and if the packet loss is more conservative, the queue length is larger. In this way, the maximum drop probability estimation method proposed in this paper is only related to the minimum threshold and queue length. When these two parameters are determined, the maximum drop probability can be calculated from the derived formula. Finally, through the NS-2(Network Simulation version 2) simulation experiment, the throughput and delay of RED routing and DropTail routing are compared by using the reset maximum drop probability value, and according to the same maximum window limit and different maximum window limitation conditions, The instantaneous queue length, average queue length and packet loss rate of the two algorithms are compared and analyzed. Finally, it is verified that the instantaneous length, average queue length, packet loss rate and the number of connections at the source end of the red algorithm are optimized after the maximum drop probability is set, so that the network congestion can be reduced and the transmission quality can be improved.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】