基于主动队列管理的RED算法及改进

发布时间：2018-06-23 19:48

本文选题：拥塞控制 + AQM　；参考：《华中师范大学》2014年硕士论文

【摘要】：随着互联网的不断发展,新型网络应用的不断涌现,特别是语音、视频等多媒体流量的增加,网络信息流量呈现爆炸式增长,带宽资源变得更加紧张,拥塞问题更加严重。另一方面,人们对于网络服务质量如传输时延、吞吐量等的要求越来越高,拥塞的产生严重影响了这些性能。在这种情况下,如何预防和控制拥塞成为了亟需解决的问题,此类问题也是国内外研究的热点。拥塞控制有两种机制,一种是基于源端的TCP协议控制,另一种则是基于中间节点(路由器)的拥塞控制。路由器不仅能够有效地监控实时队列的长度,且能够审视各个流对产生拥塞的影响,从而通知该流的源端进行调整。因此,基于中间节点(路由器)的拥塞控制机制在解决拥塞问题上有着绝对的优势。主动队列管理(Active Queue Management, AQM)是基于中间节点的拥塞控制机制中最突出的一种。本文系统的评价了几种典型的AQM算法,并对IETF推荐的唯一的AQM候选策略随机早期检测(Random Early Detection, RED)算法进行了着重介绍。RED算法解决了网络突发流量带来的问题,但是它在算法稳定度、以及参数敏感性方面仍有缺陷,因此得到了国内外广泛的研究。本文在RED算法的基础上,提出了一种非线性自适应算法(Nonlinear Adaptive Random Early Detection, NLARED),主要有两点改进：第一,在丢包概率的计算上,利用模糊数学隶属函数中的偏大型柯西分布来代替原来的线性分段函数。第二,在算法中加入自适应调整Pmax的机制。用平均队列长度来反应缓冲资源的占用情况,从而判断拥塞控制指示是否适度,对于不同的分段,引入不同基准量来调整Pmax,从而使Pmax的调整更加准确、及时,同时避免引入新的静态参数。 NS2仿真实验证明,NLARED算法能够有效地适应网络流量的变化,保持队列长度的稳定,并且减小了参数敏感性。在维持低的丢包率与高吞吐量上有着明显的提高,性能优于RED。
[Abstract]:With the continuous development of the Internet and the emergence of new network applications, especially the increase of multimedia traffic, such as voice and video, the network information traffic is explosively increasing, the bandwidth resources become more tight, and the congestion problem becomes more serious. On the other hand, the network quality of service, such as transmission delay, throughput and so on, is becoming more and more demanding, which is seriously affected by congestion. In this case, how to prevent and control congestion has become a problem that needs to be solved. There are two mechanisms for congestion control, one is based on source TCP protocol, the other is based on intermediate nodes (routers). Routers can not only effectively monitor the length of real-time queues, but also monitor the impact of each flow on congestion, thus notifying the source of the flow to adjust. Therefore, the congestion control mechanism based on intermediate nodes (routers) has an absolute advantage in solving congestion problems. Active queue Management (AQM) is one of the most prominent congestion control mechanisms based on intermediate nodes. In this paper, several typical AQM algorithms are systematically evaluated, and the random early detection (red) algorithm, which is the only candidate AQM algorithm recommended by IETF, is introduced emphatically to solve the problem caused by network burst traffic. However, it still has some defects in the stability of algorithm and parameter sensitivity, so it has been widely studied at home and abroad. Based on red algorithm, a nonlinear Adaptive Random early Detection (NLARED) algorithm is proposed in this paper. There are two main improvements: first, in the calculation of packet loss probability, The partial large Cauchy distribution in membership function of fuzzy mathematics is used to replace the original linear piecewise function. Secondly, the adaptive Pmax adjustment mechanism is added to the algorithm. The average queue length is used to reflect the occupancy of buffer resources, so as to determine whether congestion control indication is appropriate or not. For different segments, different datum is introduced to adjust Pmax, so that the adjustment of Pmax is more accurate and timely. NS2 simulation experiments show that the NLARED algorithm can effectively adapt to the network traffic change, maintain the stability of queue length, and reduce the sensitivity of the parameters. In maintaining low packet loss rate and high throughput, the performance is better than red.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】