【作者】 张杰；

【导师】 刘立芳； 杜哲；

【作者基本信息】 西安电子科技大学 ， 工程硕士（专业学位）， 2020， 硕士

【摘要】 随着网络技术和硬件设备的迅猛发展,因特网对人们的学习、工作和生活等方面产生了非常重要的影响。生活中的各种网络都是由相互联系的多个单层网络构成的,层与层之间的网络性质不同、结构不同却又相互依存,将这种复杂的网络称为多层网络。近年来网络虚拟化技术日渐成熟,研究人员提出了软件定义网络(Software Defined Network,SDN)来简化网络的管理,它是一种新型的多层网络架构。随着网络影响的不断加深,对于网络的安全性能要求也越来越高。如何迅速地检测网络状态、定位故障位置来降低故障对网络性能的影响,具有十分重要的意义。本文对基于多层网络的故障检测与诊断问题进行研究,具体分为以下三部分:(1)多层网络技术综述。通过对传统多层网络和SDN相关研究的了解,对主要的传统网络和SDN网络的故障诊断算法进行对比,并分析各算法的优点、存在的问题以及适用范围。(2)传统多层网络故障诊断算法。传统网络故障诊断算法多是针对单层网络的,相关研究已较为成熟,但基于多层网络的故障诊断研究相对较少。单层网络的故障诊断算法不能直接用于多层网络上,需要考虑层与层之间的差异性以及多层网络的特点。根据现实中多数网络仅含有少量故障的特点,本文提出了一种多层迭代更新(Malti-layer Iterative Update,MIU)故障诊断算法。首先,利用贝叶斯网络对多层网络的故障传播机制进行分析并建立多层网络故障传播模型;然后,结合网络少故障的特点,对单层网络提出启发式的可能故障假设;最后,根据网络特点和多层网络中层与层之间的关系,提出可能故障假设在多层网络中迭代的方法,并计算得出最可能故障假设。仿真结果表明本文提出的算法在少量故障的前提下比其他几种多层网络故障诊断方法在时间性能上有着巨大的提升,并在故障的诊断和误判上表现良好。(3)新型多层网络SDN故障诊断算法。目前基于SDN的故障诊断算法多需要不断地对网络进行大规模测量,并根据测量结果分析故障位置,但是这种测量会占用大量网络资源且效率低下。本文提出一种快速检测和定位的贪婪最小探测(Greedy Minimum Probes,GMP)算法。首先,采用规则管理服务器监视所有控制器并处理控制层故障;然后,通过控制器获取网络拓扑信息;其次,根据拓扑信息和头空间分析的结果,使用一种贪婪的方法得到测试集合来快速地检测网络状态;最后,结合检测结果和网络中可疑规则的数量,提出准确定位故障位置的算法。通过仿真实验,证明本文提出的算法具有更小的网络开销,能够实现快速检测网络以及准确定位故障的需求。更多还原

【Abstract】 With the rapid development of network technology and hardware facilities,the Internet has made a significant influence on people’s study,work and life.Generally,daily used networks are composed of multiple single-layer networks which are interconnected.Different layers have different network properties and structures,but they are interdependent.This complex network is called a multi-layer network.In recent years,network virtualization technology has become increasingly mature.Researchers have proposed SDN(Software Defined Network)to simplify network management.SDN is a new type of multi-layer network architecture.As the influence of the network continues to deepen,the security performance requirements of the network are becoming higher and higher.How to quickly detect the network status and locate the fault to reduce the impact of the fault on network performance is of great significance.In this thesis,fault detection and diagnosis based on multi-layer networks is studied.It includes the following three aspects:(1)Overview of multi-layer networks technology.Based on the research of the traditional multi-layer networks and SDN,the main fault diagnosis algorithms of traditional network and SDN are compared.The advantages,existing problems and applicable scope of each algorithm are analyzed.(2)Fault diagnosis algorithm of traditional multi-layer networks.The fault diagnosis algorithms of traditional network are mostly applicable to single-layer networks and the related researches are mature enough.However,researches on fault diagnosis applicable to multi-layer networks are relatively rare.The fault diagnosis algorithms of single-layer networks cannot be applied to multi-layer networks directly because of the different properties between layers and the characteristics of multi-layer networks.According to the fact that most networks only contain a small number of faults,MIU(Malti-layer Iterative Update)algorithm is proposed.First of all,the Bayesian network is used to analyze the fault propagation mechanism of the multi-layer network and a fault propagation model of the multi-layer network is established.Then,a heuristic possible fault hypothesis algorithm is proposed for the single-layer network in view of the few faults fact.Finally,based on the characteristics of the network and the relationship between the layers in the multi-layer network,a method for iterating the possible fault hypothesis in the multi-layer network is proposed,and the most probable fault hypothesis is calculated.The experimental results show that the proposed algorithm is superior to the other algorithms in time performance on the premise of the few faults,and the algorithm performs well in fault diagnosis and misdiagnosis.(3)Fault diagnosis algorithm of SDN.At present,the network shall be measured on a large scale according to most fault diagnosis algorithms of SDN,and the analysis of the location of the fault is completed in the view of the measurement results.However,such method occupies a large amount of network resources and is inefficient.GMP(Greedy Minimum Probes)algorithm with rapid detection and location is proposed in this thesis.First of all,a rule management server is used to monitor all controllers and handle control-layer faults.Afterwords,network topology information is obtained through the controllers.In the next part,a greedy method is used to obtain test sets to quickly detect the network status according to the results of the topology information and the headspace analysis.Finally,based on the detection results and the number of suspicious rules in the network,an algorithm for accurately locating faults is proposed.Through simulation experiments,it is proved that the algorithm proposed in this thesis has smaller network overhead and can meet the requirements of rapid network detection and accurate fault location.更多还原