NFV环境下级联故障分析及解决方案研究

发布时间：2019-04-15 18:41

【摘要】：随着互联网的发展,上层服务需求越来越多样化,传统的底层网络功能迫于专属硬件的限制,无法快速更新适应。网络功能虚拟化(Network Function Virtualization,NFV)将IT虚拟化技术引入传统电信网络,实现了传统网络功能的软硬件分离,使得网络功能可编程动态伸缩,大大增强了网络的更新迭代速度。然而,传统电信网络和IT网络在可靠性的要求上存在较大差异,级联故障问题频繁发生在IT网络系统中,针对电信网络的高可靠性要求,NFV环境下的网络的级联故障问题应当引起重视。本文首先从服务功能链(Service Function Chain,SFC)的角度分析了NFV环境下故障扩散的模式,通过实验总结出两种不同的故障扩散模式:(1)当不同服务功能链共享的资源故障时,故障将在大量不同服务链间扩散。(2)当服务功能链上不同功能模块间性能不匹配时,故障会沿着服务功能链的性能瓶颈模块向上游模块扩散。为了防止共享资源故障引起的大面积故障扩散,本文采用对关键资源如共享资源加强保护的办法提高系统的可靠性,阻断故障的扩散。首先针对NFV多层虚拟化特点,对其不同层次中各模块之间的存在的多种关系(依赖、备份、负载均衡)进行抽象,建立了一个三值状态多层依赖模型。然后提出一个优化系统可靠性的备份资源分配算法,仿真结果表明,算法能得到最优解或近似最优解。从而从备份资源分配角度为实现故障隔离提供思路。为了解决服务链上不同服务功能模块服务能力不匹配带来的扩容风暴问题,本文提出了联动扩容算法。该算法主要思想在于预测出瓶颈模块扩容后对下游模块的影响,将受影响的模块与瓶颈模块联动扩容,从而避免多次扩容。具体方法是利用神经网络,建立了扩容量和数据流的输入输出速率之间的关系模型,使得上游节点的扩容对下游节点的影响可以量化,提前对下游受波及的节点进行扩容,防止了故障的扩散。仿真结果表明,联动扩容算法避免了单点独立扩容可能导致的二次服务功能不匹配问题,节约了系统扩容的总时间,降低了平均服务响应延迟。最后,为更好地应用上述算法,在目前业界的NFV架构基础上,实现了NFV环境下故障隔离架构,该架构根据上述算法的输入输出需求,增加了监视器模块和算法部署模块,并给出了运行实例,为增强算法的实用价值提供参考。
[Abstract]:With the development of Internet, the demand of upper layer service is becoming more and more diversified. The traditional bottom network function is forced by the limitation of exclusive hardware, and can not be updated and adapted quickly. Network function Virtualization (Network Function Virtualization,NFV (Network function Virtualization) introduces IT virtualization technology into traditional telecommunication network, realizes the separation of hardware and software of traditional network function, makes the network function dynamically and programmatically scalable, and greatly enhances the iterative speed of network update. However, there is a great difference between the traditional telecommunication network and the IT network in the reliability requirements. Cascading failures occur frequently in the IT network system, aiming at the high reliability requirements of the telecommunication network. The cascading fault of network in NFV environment should be paid more attention. In this paper, the mode of fault diffusion in NFV environment is analyzed from the point of view of service function chain (Service Function Chain,SFC), and two different modes of fault diffusion are summarized by experiment: (1) when the resource shared by different service function chain fails, The fault will spread among a large number of different service chains. (2) when the performance of different functional modules on the service function chain does not match, the fault will spread along the performance bottleneck module of the service function chain to the upstream module. In order to prevent the large-scale fault diffusion caused by shared resource fault, this paper adopts the method of strengthening the protection of key resources such as shared resource to improve the reliability of the system, and to block the spread of fault. According to the characteristics of NFV multi-layer virtualization, this paper abstracts the multi-layer relationships (dependency, backup, load balancing) between the modules in different levels, and establishes a three-valued state multi-layer dependency model. Then a backup resource allocation algorithm is proposed to optimize the reliability of the system. The simulation results show that the algorithm can obtain the optimal solution or approximate optimal solution. From the point of view of backup resource allocation, this paper provides a train of thought for the realization of fault isolation. In order to solve the problem of capacity expansion caused by the mismatch of service capability of different service function modules in service chain, this paper proposes a linkage capacity expansion algorithm. The main idea of this algorithm is to predict the impact of bottleneck module expansion on downstream modules, and to link the affected modules with bottleneck modules so as to avoid multiple capacity expansion. The specific method is to use the neural network to establish the relationship model between the expansion capacity and the input / output rate of the data stream, so that the influence of the upstream node expansion on the downstream nodes can be quantified, and the affected nodes in the downstream nodes can be expanded in advance. The failure was prevented from spreading. The simulation results show that the linkage expansion algorithm avoids the mismatch problem of secondary service function which may be caused by single point independent capacity expansion, saves the total time of system expansion, and reduces the average service response delay. Finally, in order to better apply the above algorithms, based on the current industry NFV architecture, the fault isolation architecture under the NFV environment is implemented. According to the input and output requirements of the above algorithms, the monitor module and the algorithm deployment module are added. A running example is given to provide a reference for the practical value of the enhancement algorithm.
【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】：TN915.0

【参考文献】