NFV Oriented Edge Computing Anomaly Detection And Root Cause Analysis

In recent years,the number of mobile terminal devices has grown rapidly,but the computing power and storage resources of mobile terminal devices are limited,which cannot meet the needs of computing intensive and delay sensitive applications.In order to improve user quality of service and reduce equipment deployment costs and operating costs of operators,the integration of edge computing and network function virtualization has become a feasible solution.Deployment of virtual network functional components in base stations of fog computing mobile wireless access network provides mobile users with flexible and high-quality services,however,it also causes the complexity of network layout and management,and one of the main challenges is anomaly detection and root cause analysis in fault management.Therefore,anomaly detection and root cause analysis in network function virtualization(NFV)oriented edge computing scenario has become a hot research direction.In the context of fog computing mobile wireless access network,this thesis studies the anomaly detection and root cause analysis in NFV oriented edge computing scenario.The research work and innovations mainly include:(1)Considering the correlation between network topology structure and virtualized network function(VNF)deployment,anomaly detection and root cause analysis is formulated,and a disturbance-based anomaly detection and root cause analysis method is proposed,which include anomaly detection algorithm based on graph node,edge message interaction network and anomaly root cause analysis algorithm based on mutual information and counterfactual.The effectiveness of the proposed algorithms are verified by simulation.(2)Considering the dynamic characteristics of networking and computing resources,an anomaly detection algorithm based on subgraph message passing and feature extraction network is proposed by using graph neural network self-interpretation method.The proposed algorithm generates interpretation results when generating detection results.The interpretable process of generating the anomaly detection results is given effectively.Compared with the baseline algorithms,simulation results show that the proposed algorithm has improved performance in terms of accuracy,true positive rate,false positive rate,sparsity,etc.