Resource Optimization Based On Graph Learning In Fog Enabled Radio Access Network

Recently,the number of mobile devices experience explosive increase,but most mobile devices have limited computing capacity and storage resources,making it difficult to support computation intensive services.Fog computing supports the deployment of computing resources and storage resources in wireless access network,and can provide computation intensive services based on cloud-side collaboration for mobile terminals.It can not only effectively reduce user service delay and ensure service quality experience,but also effectively alleviate the backhaul link load traffic burden of the core network.The allocation strategy of computing resources,storage resources and link bandwidth resources on fog nodes has an extremely important impact on the support of computing-intensive services in mobile wireless access networks.Due to the dynamic change of network load in mobile wireless access network,how to realize flexible and reasonable resource allocation in dynamic network is a challenge.This thesis focuses on the optimization of VNF resource optimization problem in fog enabled radio access networks.The main research work and innovations include the following two aspects:(1)In the scenario of fog enabled radio access network,considering dynamic network load,the SFC routing problem is formulated as an optimization problem which aims to minimize the end-to-end delay of SFCs.Based on spatiotemporal graph convolutional neural network,an algorithm called STGRA is proposed.The effectiveness of STGRA algorithm is verified by simulation.The simulation results also show that STGRA has got the tradeoff between the performance and computation complexity.(2)Considering the dynamics of network load and networking topology of the optimization problem on VNF placement in fog enabled radio access network,based on graph convolutional network,a VNF placement optimization algorithm is proposed based on graph learning called VP-GL.In order to deal with the problem of labelling data when training datasets for VP-GL,a model training based labelling method is proposed,which is based on the index of affinity between nodes and VNFs.The performance of VP-GL algorithm is evaluated by simulation.The effectiveness of the index of affinity for training labels is verified.The simulation results show that the VP-GL algorithm has good performances in terms of average network service delay and SFC service request acceptance ratio.Evaluation results also reveal that VP-GL algorithm can not only solve the optimization problem of deploying VNFs dynamically,but also adapt to dynamic changes in network load and network topology.