基于动态时间扭曲的NFV网络故障检测  

Fault detection based on dynamic time warpping in NFV networks

作　　者：齐小刚 张皓然 李家慧 单明媚 Qi Xiaogang;Zhang Haoran;Li Jiahui;Shan Mingmei(School of Mathematics and Statistics,Xidian University,Xi'an,710071,China)

机构地区：[1]西安电子科技大学数学与统计学院,西安710071

出　　处：《南京大学学报（自然科学版）》2023年第2期343-350,共8页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61877067)。

摘　　要：网络功能虚拟化(Network Functions Virtualization,NFV)实现了硬件与软件的解耦,是5G应用与发展的核心技术.NFV网络具有高动态性、层间不可见性、网络组件复杂等特点,给网络管理带来了挑战.为了保证NFV网络的正常高效运行,提出基于动态阈值的NFV故障检测算法DTF-DTW:首先,根据已有的正常样本数据,获得不同虚拟网络功能的故障阈值;然后,采用基于滑动窗口的快速动态时间扭曲算法,检测虚拟网络功能的故障时段并报警.采用真实世界的NFV数据对该算法进行测试,并使用机器学习算法STAD (Spatio-Temporal Anomaly Detection)进行比较,证明提出的算法与现有算法相比,能更加迅速准确地检测出NFV中的异常行为,提高网络运行效率和用户服务满意度.Network Function Virtualization(NFV)decouples hardware and software,which is regarded as the core technology for the application and development of 5G.NFV is characterized by high dynamics,invisibility between layers,and complexity of network components,which bring challenges to network management.Due to ensure the normal and efficient operation of NFV networks,this paper presents Dynamic Threshold-based Dynamic Time Warping(DTF-DTW)algorithm for fault diagnosis.Based on the sliding window,DTF-DTW retains the temporal characteristics of data,and detects the fault periods for different Virtual Network Functions(VNFs)by Dynamic Time Warping(DTW)method.The proposed DTF-DTW algorithm is tested using real-world NFV environment data and the simulation results shows that our algorithm identifies the abnormal behaviors in NFV networks more efficiently than existing algorithms,improving the network operation efficiency and user service satisfaction.

关 键 词：NFV 网络故障检测 动态时间扭曲 时间序列 滑动窗口 动态阈值 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]