Proactive SFC Orchestration Algorithm Based On Traffic Flow And Resource State Prediction

With the booming development of new applications in the mobile Internet of Things,the conflict between the limited storage and computing capacity of mobile devices and the surging computing demand of applications is becoming increasingly obvious.Mobile Edge Cloud(MEC)has become an effective technology to resolve the conflict between the two.In the MEC environment,the virtual resource allocation method and the deployment of virtual network function(VNF)on the dedicated hardware infrastructure can effectively reduce the service cost.However,the existing virtual resource allocation methods,including VNF resource mapping and migration,are mostly passive and can not respond to the service need of users immediately.Therefore,it is of great significance to study the active VNF resource allocation method in MEC network.To solve the problem of service response delay in most passive virtual resource mapping methods,a VNF dynamic scaling and deployment algorithm based on traffic prediction is proposed.In this method,a traffic prediction model of the service chain based on the Gated Recycle Unit(GRU)was established firstly,and a VNF elastic scaling strategy was designed based on the traffic prediction value.The horizontal scaling of the VNF was used to avoid service interruption.Furthermore,based on the extended VNF instances and considering the physical node capacity and other constraints,a VNF instance deployment algorithm based on deep reinforcement learning is designed to minimize the guarantee cost.Finally,according to the deployment location of each VNF instance,the traffic routing algorithm of the service chain based on business priority is designed to ensure the quality of service of high-priority business and minimize the cost.The simulation results show that the proposed active VNF deployment algorithm can reduce the profit loss caused by service response delay,and the proposed VNF deployment algorithm can save 7.4%and 22.2%cost compared with other reinforcement learning deployment algorithms and greedy algorithms,respectively.In order to ensure the normal operation of the Service Chain(SFC)and not be affected by the failure of physical nodes,this paper proposes a service migration algorithm based on fault prediction to solve the hysteretic problem of the current VNF fault protection mechanism.Firstly,a physical node fault prediction model is constructed to predict whether each node in the network will fail in the future,and then a cost-and business-priorities-oriented VNF instance migration algorithm is designed based on the prediction results.The simulation results show that the proposed method can effectively guarantee the normal operation of 95%SFC,and the proposed VNF instance migration algorithm can save 21.3%and 26.6%cost compared with the heuristic algorithm and the greedy algorithm.