基于单播的网络拓扑推断的准确度研究

发布时间：2018-04-23 00:21

本文选题：网络拓扑测量 + 网络断层扫描技术　；参考：《兰州交通大学》2014年硕士论文

【摘要】：随着互联网技术深入到人们社会生活的各个行业领域，如金融行业、通信服务行业、煤炭行业、石油行业、食品流通行业等，人们的日常生活与互联网的关系日益密切，与此同时，对网络实时、准确且可靠的传输服务的要求也越来越高，尤其是网络传输延时和网络安全方面。网络测量作为网络管理和监控的依据，显得越来越重要。拓扑结构，是网络测量的重要方面，是一切互联网活动顺利开展的保障。网络断层扫描技术(Network Tomography，NT)是一种基于端到端的测量技术，该技术通过从源节点发送探测包到特定的目标节点，并在目标节点统计到达的探测包的各种参数信息(丢包率、时延、时延抖动和链路带宽等)，根据这些信息来推断节点间的兄弟关系。NT技术有效克服了传统网络测量的缺点，是现阶段推断网络拓扑的最有效技术。基于NT技术的拓扑测量依据数据信息采集方式的不同可分主动测量和被动测量，主动测量又分为基于多播的测量和基于单播的测量两种方式。由于设备是否支持多播功能与网络安全性无直接联系，所以现有网络中的设备并不都支持多播功能，从而使得多播测量方法的适用范围受到限制。而基于单播的测量方法对网络设备本身的要求不高，有更广的适用性。本文的研究将基于单播测量方式展开，主要内容如下：首先概要地阐述了网络拓扑测量技术的背景、意义及国内外现状，并对传统网络测量技术和NT技术分别进行了概述，比较分析了传统网络测量方法进行拓扑推断的优势及其存在的不足。其次论述了端到端的测量的两种方式：主动测量和被动测量，以及基于主动测量的多播测量方法和单播测量方法，并对基于单播测量的几种探测包模型进行了论述，分析比较了它们各自的优缺点。然后论述了三元分组列车测量模型及现有的基于时延和基于时延抖动的两种三元分组列车测量方法。在现有基于单参数(时延或时延抖动)的三元分组列车测量方法基础上，提出一种利用双参数来推断网络拓扑结构的三元分组列车测量方法，系统阐述了此方法进行网络拓扑识别的测量过程，并在Ubuntu下的NS2仿真平台上模拟探测包的发送过程，对相关数据进行统计。实验在不同的网络负载情况下对现有方法与改进方法的拓扑推断准确度进行比较，结果验证了该方法的有效性与准确性。最后论述了将此改进方法应用于推断已知拓扑中添加新节点之后的拓扑结构，并在NS2下进行实验，验证了其准确性。该方法在推断过程中引入了叶节点的高度，故在推断叶节点层次较为复杂的拓扑时，，该方法有效减少了探测包的发送量。
[Abstract]:With the development of Internet technology in various fields of people's social life, such as finance, communication service, coal, petroleum, food circulation and so on, the relationship between people's daily life and the Internet is becoming more and more close. At the same time, the demand for real-time, accurate and reliable transmission service is becoming higher and higher, especially in the aspect of network transmission delay and network security. As the basis of network management and monitoring, network measurement is becoming more and more important. Topology, an important aspect of network measurement, is the guarantee of the smooth development of all Internet activities. Network TomographyNT-is an end-to-end measurement technique, which sends detection packets from source nodes to a specific target node, and counts the various parameters (packet loss rate, delay) at the target node. Time delay jitter and link bandwidth are used to infer the sibling relationship between nodes. NT technology overcomes the shortcomings of traditional network measurement and is the most effective technique to infer network topology at this stage. The topology measurement based on NT technology can be divided into active measurement and passive measurement according to the different ways of collecting data information. Active measurement can be divided into two ways: multicast based measurement and unicast based measurement. Because there is no direct relation between the multicast function and the network security, not all the devices in the existing network support multicast function, which limits the scope of application of multicast measurement method. However, the measurement method based on unicast does not require the network equipment itself, so it has wider applicability. The research of this paper will be based on unicast measurement, the main contents are as follows: Firstly, the background, significance and current situation of network topology measurement technology are briefly described, and the traditional network measurement technology and NT technology are summarized respectively. The advantages and disadvantages of traditional network measurement methods for topology inference are compared and analyzed. Secondly, two methods of end-to-end measurement: active measurement and passive measurement, multicast measurement based on active measurement and unicast measurement are discussed, and several detection packet models based on unicast measurement are discussed. Their advantages and disadvantages are analyzed and compared. Then, the measurement model and two existing measurement methods based on delay and delay jitter are discussed. On the basis of the existing measurement methods based on single parameter (delay or delay jitter), this paper proposes a new method for the measurement of ternary packet trains, which uses two parameters to infer the topological structure of the network. The measurement process of network topology identification by this method is described systematically, and the sending process of probe packets is simulated on the NS2 simulation platform under Ubuntu, and the relevant data are statistically analyzed. The experimental results show that the proposed method is effective and accurate by comparing the existing methods with the improved methods under different network load conditions. Finally, this improved method is applied to infer the topology structure after adding new nodes to the known topology, and the experiment is carried out under NS2 to verify its accuracy. The height of the leaf node is introduced in the process of inference, so the method can effectively reduce the transmission of the probe packet when inferring the complex topology of the leaf node.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.02

【参考文献】