【作者】 刘岩；

【导师】 刘恩海；

【作者基本信息】 河北工业大学 ， 计算机应用技术， 2011， 硕士

【摘要】 在网络迅速发展的当今社会,网络的使用者要求网络提供高速度、高质量的信息传输服务,与此同时,网络拥塞的现象却屡屡发生。因此,拥塞控制的研究也成为了研究者青睐的研究方向。路由器缓存中存在过多的等待发送的数据包,网络的带宽容量却又不能承受如此之大的负荷,这就会造成拥塞现象。解决网络拥塞的核心就是队列管理和队列调度算法的实现,队列管理算法是解决路由器内部队列如何建立、如何维护、如何排队的过程,队列调度算法是用来决定谁先被调度的算法,以此来实现队列之间共享输出链路资源的过程。本文是在RED拥塞控制算法的基础上进行研究的,RED算法是队列管理算法中的一个经典代表,属于主动队列管理算法的范畴。通过对RED算法优缺点进行详细分析,提出了一种基于流量预测的改进RED算法——2P-RED算法。在研究思路上,首先针对网络流量的自相似性、长相关性、周期性等特性,利用数学公式对流量特性进行量化,为建立预测模型提供了基础；其次研究了各种智能算法,提出了把人工神经网络模型应用到数据流量的预测的想法,用Matlab工具进行仿真实验,为了提高BP算法的精确度和学习能力,BP神经网络中权值阈值的初始化利用模拟退火和粒子群算法进行了改进；然后,将流量预测代码添加到RED协议当中去,实现对RED算法的改进,添加协议的过程主要工作是对Edv结构体以及类REDQueue中drop_early函数进行修改,协议修改完毕,在NS2模拟软件中重新编译,即可投入到路由器队列管理算法的使用当中了。文章最后建立了含有不同个数的TCP、UDP数据流的网络模型,数据包传送过程分别采用改进的RED和基本RED两种队列管理算法,由模拟得到的Trace文件可以进一步分析出不同算法的丢包率、吞吐量、时延来,实验结果验证了改进算法在解决拥塞控制上具有良好的效果。更多还原

【Abstract】 With the rapid development of network times, network users request the network must provide high-speed, high-quality services. At the same time, the network congestion often occurs. Therefore, congestion control research has become a popular research area.It will cause the congestion, because there are many packets waiting to be sent in the cache of router, but the network bandwidth capacity cannot bear such a large load. It is the queue management algorithm and the queue scheduling algorithm that to solve the congestion. The queue management is the process to solve how to create, maintain and line up queue, and the queue scheduling algorithm is to determine which should be scheduling algorithm, and in order to achieve sharing the link of resources between output queues.The Random Early Detection (RED) algorithm is a classical representative of the active queue management, of which study on the basis. Through a detailed analysis about advantages and disadvantages of RED algorithms, this paper put forward the 2P-RED algorithm, which is an improved RED algorithms based on flow prediction.In the study, first, put these characterisitics to formula according to the nature of self-similarity, long-range dependence, periodicity of the network traffic. It is the foundation for the establishment of forcasting model.Second, research the intelligent algorithms and put forward the idea of applying the artificial network model into the data flow prediction, then simulate with Matlab.The weight threshold initialized using SA and PSO algorithm, after that the accuracy and learning ability of the BP algorithm are improved. And then, it adds the flow prediction code to RED protocol to improve RED algorithm. The main task of adding the agreement is amending the function of drop-early in the RED Queue and Edv Structure. After amending this agreement, modified and re-compiled in the Network Simulator 2(NS2) simulation software, it can be put into the router queue management algorithm and used.Finally, the paper establishes the network model which includes different number of TCP, UDP data flow. Comparing the basic RED algorithm and the improved RED algorithm, and get the packet loss rate, throughput and delay by analyzing the trace files, the results show the improved algorithm has a good effect in solving the congestion control.更多还原