Research On Fault Detection And Diagnosis Algorithm Of Service Function Chain In Network Slicing Scenario

In order to meet people's diverse needs for communications services,China will achieve the large-scale deployment of 5G networks in 2020.The introduction of network slicing technology in 5G networks effectively solves the problems of rigidity and closure of the existing network architecture,which greatly improves the flexibility of the network and the utilization of resources.However,network slicing technology not only realizes flexible network configuration and resource sharing,but also brings greater challenges to network management and operation.Therefore,how to realize the automation and intelligence of network fault management is the key to ensure the stable operation of the network.This thesis focuses on the fault detection and diagnosis algorithm of the service function chain in the network slicing scenario,and the main research contents and innovations of the thesis are summarized as follows:1.Aiming at the problem that the fault of the service function chain can be propagated in different virtual network function nodes of the service function chain in the5 G network slicing scenario,firstly,a scheme is designed to monitor the working state of each virtual network function node in the service function chain and collect relevant performance data at the application layer,and detect the fault by analyzing the data.Secondly,considering the timing,multi-source and high-dimensional characteristics of the network monitoring data,as well as the proactive requirements of fault detection,a fault detection model based on Gated Recurrent Unit(GRU)network prediction is proposed,and the network model parameters are trained with historical data sets to predict the working state of the network.Finally,considering the problem that the lack of data in the service function chain and it takes a long time to model with the historical data,which is not conducive to the real-time requirement of fault detection,transfer learning is introduced to accelerate the convergence rate of GRU network learning model.The simulation results show that the proposed fault detection algorithm can converge in a short time while ensuring the detection accuracy.2.Aiming at the problem that the fault of the underlying physical node will cause abnormal performance of multiple service function chains running on it in the 5G network slicing scenario,firstly,according to the multi-layer propagation relationship of the fault in the network virtualization environment,a dependency graph model of faults andsymptoms is constructed,and the symptom data is collected by monitoring performance data of multiple virtual network functions on physical nodes.Secondly,considering the diversity of network symptom observation data under the network slicing architecture and the spatial correlation between the physical node and the function of the virtual network,the deep belief network is introduced to extract the characteristics of the observation data.Finally,dynamic Bayesian network is introduced to diagnose the root cause of faults in real time by using the temporal correlation between faults.The simulation results show that the proposed fault diagnosis algorithm can effectively diagnose the root cause of faults and has good diagnostic accuracy.