【作者】 罗维东；

【导师】 刘兴伟；

【作者基本信息】 西华大学 ， 计算机软件与理论， 2008， 硕士

【摘要】 随着Internent的快速发展,人们的生活方式和传统的信息交换方式也受到巨大的影响。伴随着各种各样的应用需求和网络规模越来越大,网络管理工作也越来越繁重和困难。网络拥塞、网络安全、网络故障等一系列问题时刻影响着信息社会的正常发展。网络流量的测量和预测对大规模网络管理、规划和设计具有非常重要的意义。本论文主要研究内容和特色如下:(1)通过查阅国内外相关文献资料,对网络流量测量方法进行研究。常用的3种测量技术是:SNMP测量、Packet Sniffing测量和Netflow测量。SNMP是通过提取网络设备(代理)提供的MIB(管理对象信息库)中收集的一些具体设备以及流量信息有关的变量进行测量的。Packet Sniffing是一种用网卡在链路层来获取网络流量的方法,使用时将它串接在需要捕捉流量的链路中,通过分流链路上的数字信号来获取流量信息。Netflow是一种数据交换方式,流量采集是基于网络设备提供的Netflow机制实现的。(2)介绍常用预测方法。如最小均方误差、神经网络、ARMA模型以及基于统计学习理论的支持向量机等。(3)通过NS2仿真器,结合支持向量机回归理论对模拟的网络流量进行预测实验。首先分析了NS2软件包的结构、仿真原理和流程,然后利用NS2模拟简单网络拓扑结构并产生流量数据,为了提高预测精度将原始数据进行差分处理,然后组织训练数据和检验数据,最后基于SVM建立预测模型,用训练数据训练模型参数,然后用检验数据进行流量预测,并分析结果,结果表明实验预测效果比较理想。(4)在仿真实验可行的基础上,采用Netflow技术对真实环境下的网络流量进行测量,并进行预测分析。在某网管中心的核心路由器上进行Netflow配置,并在一台PC机上通过Netflow相关工具进行Netflow包收集和流量计算。最后对真实数据进行预测分析,结果表明SVM预测方法对高非线性、抖动剧烈的流量数据预测效果较好。更多还原

【Abstract】 With the fast development of Internet, people’s living style and methods of traditional information exchange have been affectted seriously. Various application requirements and Internet scale-up lead to much more difficulties on the network management. Network congestion, security and malfunction etc. have influence on the normal development of the information society. Internet traffic measure and forecast have huge significance on the management, layout and design of large-sized network. Nowadays, Netflow method provided by Cisco is used comprehensively and likely to be the factual standard on measure. Support Vector Machine, a newly machine learning algorithm, having some unique merits such as rapidness of solution and strong generalization performance, shows good performance on nonlinear regression.The main research contents and feature of this paper are as follows:(1) This paper introduces the traffic measure methods in common used. SNMP measure collects traffic through the variants in the MIB provided by SNMP network agents. Packet sniffing measure uses NIC (Network Interface Card) to get the traffic at the Link layer. And netflow measure gets traffic by the data exchange mechanisim.(2) The paper introduces some forecast methods in common used. Such as Min-average error method, Artificial Nerual Network method, Self-regression glide average model method and ARMA model method etc. The paper mainly studys the SVM (Support Vector Machine) method based on the statistics theory.(3) The paper introduces the structure of NS2 software pack, NS2 simulation principle and process. Generate traffic data by NS2 simple network topology. Make difference operation on those data and orgnize training data and checkout data. Build SVM forecasting model, get model parameters by traning data, and verify the forecast accuracy by the checkout data. The result on the precision of forecast is good and feasible.(4) Based on the feasible of simulation, using Netflow measure to get traffic in the real circumstance, and forecasting. Through making Netflow configuration on router of Network Mangement Center, collects Netflow packets and computes traffic by related Netflow tools on an appointed PC. Forecasting on that traffic, the result shows SVM regression method having good effect even if the traffic is high nonlinear and huge oscillation.更多还原