网络拥塞控制的主动队列管理的研究

发布时间：2018-03-30 21:08

本文选题：主动队列管理　切入点：自适应　出处：《南京邮电大学》2014年硕士论文

【摘要】：随着以互联网（Internet）为代表的现代通信网络的迅速发展，越来越多的用户通过Internet传递和获取信息，用户的需求日趋丰富和多样化，端到端拥塞控制机制已经很难避免或者缓解网络拥塞现象的出现。因此，研究网络内部的拥塞控制技术是网络管理中需要解决的问题。主动队列管理（Active Queue Management，AQM）作为网络中间的路由节点中重要的拥塞控制技术，正逐渐成为网络拥塞控制领域研究的热点。目前，在路由节点中实施AQM策略辅助基于传输控制协议（TransmissionControlProtocol，TCP）的拥塞控制机制，是解决网络拥塞问题的一个主要途径。本文主要研究通信网络模型下路由节点中的AQM算法。首先详细地描述网络拥塞控制原理，然后详尽地阐述AQM算法的组成、设计规则、性能指标、公平性，接着进一步分析几种经典的AQM算法，并且对比例积分微分（Proportional Integral Derivative，PID）控制器进行优化，提出一种改进的自适应PID神经网络（MAPIDNN）算法。MAPIDNN算法引入采用附加动量因子的梯度下降法调整权值的PID神经网络控制器，，并且在每个分组到达时刻利用二次方函数自适应地调整丢包率来适应实际队列长度的变化，同时引入无状态流量匹配机制增强TCP适应流的生存能力。仿真实验结果表明，与PID、PIDNN算法相比，MAPIDNN算法具有更好的暂态性能，较高的吞吐率和瓶颈链路利用率，最小的分组排队时延及时延抖动，同时增强了TCP适应流的生存能力。
[Abstract]:With the rapid development of modern communication network represented by Internet, more and more users transmit and obtain information through Internet. End-to-end congestion control mechanism has been difficult to avoid or alleviate the phenomenon of network congestion. It is an important problem in network management to study the congestion control technology in network. Active Queue Management (AQM) is an important congestion control technology in the routing nodes in the middle of the network. At present, the implementation of AQM strategy to assist the congestion control mechanism based on Transmission Control Protocol (TCP) is one of the main ways to solve the network congestion problem. This paper mainly studies the AQM algorithm in the routing nodes under the communication network model. Firstly, the principle of network congestion control is described in detail, then the composition, design rules, performance index, fairness of the AQM algorithm are described in detail. Then, several classical AQM algorithms are analyzed, and the proportional Integral derivative controller is optimized. An improved adaptive PID neural network MAPIDNN algorithm .MAPIDNN algorithm is proposed to introduce the PID neural network controller which uses gradient descent method with momentum factor to adjust weights. At each packet arrival time, the quadratic function is used to adaptively adjust the packet loss rate to adapt to the change of actual queue length, and the stateless traffic matching mechanism is introduced to enhance the survivability of TCP adaptive flow. Compared with PID-PIDNN algorithm, MAPIDNN algorithm has better transient performance, higher throughput and bottleneck link utilization, minimum packet queuing delay delay and delay jitter, and enhances the survivability of TCP adaptive flow.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.07;TP183

【相似文献】