网络功能虚拟化环境下的故障管理

发布时间：2018-06-08 18:22

本文选题：网络功能虚拟化 + 故障检测　；参考：《电子科技大学》2017年硕士论文

【摘要】：为了满足日益增长的网络服务需求,越来越多的网络中间件被部署到网络中使得网络僵化问题严重。网络功能虚拟化从运营商的角度对现有网络架构进行改造,通过软硬件资源解耦,将各种网络硬件设备整合到行业标准的x86服务器中并使用软件实现的网络功能来替代专有设备的功能,有效提升了网络功能的开发效率,也使得网络功能的部署和管理更加灵活方便。自提出以来,网络功能虚拟化已在学术界和工业界引起了广泛的关注。然而网络功能虚拟化在提升效率的同时,也给网络可靠性带来一些挑战。虚拟化的过程占用额外的系统资源,而且虚拟节点相较于传统物理节点更加复杂。这些因素导致网络功能虚拟化环境下具有更高的故障发生概率。故障发生后,能否快速准确地进行故障检测和定位对于维护系统长期稳定可靠地提供服务至关重要。故障管理是网络管理的一项基本内容,其目标是尽快消除故障并恢复服务的正常运行,以满足和用户达成的服务等级协议的要求。本文对网络功能虚拟化环境下的故障检测和故障定位等两项核心技术进行了研究。故障检测负责捕获系统发生的故障,而故障定位负责对故障的原因和位置进行定位。针对网络功能虚拟化的环境特性以及传统数值检测技术的缺陷,本文基于无监督机器学习的方法构建故障检测模型。提出了一种基于模型融合的无监督故障检测机制,该机制可以有效覆盖传统检测算法的检测盲区并提高检测性能。此外,本文还针对突发特性造成的错误告警现象进行处理,提出了基于时间窗的修正机制过滤错误告警。在故障定位方面,本文对网络功能虚拟化环境下故障原因定位和故障位置定位的方法进行了研究,提出了一种基于相关性分析的分层故障定位机制。该机制基于时间和特征变异程度的相关性分析对故障原因进行定位,基于时间、服务功能链和资源共享等方面的相关性分析对故障位置进行定位。通过拓展依赖关系图增加服务功能链引入的相关性,可有效提高定位结果的准确性。定位机制通过“自顶向下,逐层定位”的实施原则集成各个层次的定位结果,突破了信息获取受限的瓶颈。本文以三层云服务为基础搭建仿真环境,并向其中注入几种不同类型故障进行实验,仿真结果显示本文提出的解决方案可以有效检测并定位注入的故障。
[Abstract]:In order to meet the increasing demand for network services, more and more network middleware is deployed to the network, which makes the network rigid. Network function virtualization transforms the existing network architecture from the point of view of the operator and decouples it through software and hardware resources. Integrating all kinds of network hardware devices into the industry standard x86 server and using the network function realized by the software to replace the function of the proprietary equipment, the development efficiency of the network function is improved effectively. It also makes the deployment and management of network functions more flexible and convenient. Since it was put forward, network function virtualization has attracted wide attention in academia and industry. However, network function virtualization not only improves efficiency, but also brings some challenges to network reliability. The virtualization process takes up additional system resources, and virtual nodes are more complex than traditional physical nodes. These factors lead to higher probability of failure in virtualization of network function. After the fault occurs, it is very important to quickly and accurately detect and locate the fault for the maintenance system to provide service stably and reliably for a long time. Fault management is one of the basic contents of network management. Its goal is to eliminate the fault as soon as possible and resume the normal operation of the service to meet the requirements of the service level agreement reached with the user. In this paper, two core technologies, such as fault detection and fault location, are studied in network function virtualization environment. Fault detection is responsible for capturing the fault of the system, and fault location is responsible for locating the cause and location of the fault. Aiming at the environmental characteristics of network function virtualization and the shortcomings of traditional numerical detection techniques, this paper constructs a fault detection model based on unsupervised machine learning. An unsupervised fault detection mechanism based on model fusion is proposed, which can effectively cover the blind areas of traditional detection algorithms and improve detection performance. In addition, this paper also deals with the error alarm caused by burst characteristics, and proposes a time window based correction mechanism to filter error alarm. In the aspect of fault location, this paper studies the methods of fault cause location and fault location in network function virtualization environment, and proposes a hierarchical fault location mechanism based on correlation analysis. The mechanism is based on the correlation analysis of time and degree of feature variation to locate the fault cause, and based on the correlation analysis of time, service function chain and resource sharing to locate the fault location. By extending the dependency graph to increase the correlation introduced by the service function chain, the accuracy of the location results can be improved effectively. The localization mechanism integrates the localization results of different levels through the implementation principle of "top-down, location-by-layer", which breaks through the bottleneck of limited access to information. This paper builds a simulation environment based on the three-layer cloud service, and injects several different types of faults into the simulation environment. The simulation results show that the proposed solution can effectively detect and locate the injected faults.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【参考文献】