Research On Energy Saving Strategy Of Network Functions Virtualization Architecture Based On Resource Prediction

In Network Functions Virtualization(NFV),Network functions implemented as software on general purpose servers are called Virtualized Network Function(VNF).The feature of virtualization enables dynamic migration of Service Function Chaining(SFC),which provides necessary flexibility for services.As VNF resource requirements change with network traffic,there may be a large number of underutilized servers in the network,resulting in huge energy consumption.Therefore,how to minimize the long-term network energy consumption while ensuring the Service Level Agreement(SLA)is a very important issue in NFV networks.In the past,passive integrated VNF energy saving strategies lack research on migration timing.In order to eliminate the lag of VNF migration,effective resource prediction for VNF is a very important solution.This thesis mainly studies the energy-saving migration strategy of network functions in NFV networks.In view of the SFC service due to resource requirements in the process of dynamic change of network energy consumption problems,this paper proposes a virtual network function based on resource prediction energy-saving dynamic migration algorithm,the algorithm takes advantage of the network resources demand forecasting results,eliminate dynamic migration lag,and according to the network traffic and server resource usage of each time determine the migration time,by timely splitting VNF servers that may be overloaded to eliminate SLA violations,and timely merging low-load servers to save longterm network energy consumption.At the same time,the objective power consumption of the server is taken into account to avoid frequent switching of the server.Simulation results show that compared with existing algorithms,RP-EDM has good effect and stability under different prediction results with different precision.Under the condition of ensuring lower SLA interruption rate,RPEDM can improve the total traffic of network services by more than 15%and reduce the energy consumption cost by more than 20%.To solve the problem of virtual network function resource prediction,this thesis proposes a multi-stage service function chain migration architecture based on resource prediction,which optimizes network energy consumption through offline pre-training,online migration and online incremental learning.This architecture includes a serial-to-sequence resource prediction model based on attention mechanism,which learns the connection dependencies between VNFs in SFC and adopts a serial-to-sequence structure to make the model predict multiple VNF simultaneously.At the same time,an SFC deployment algorithm based on particle swarm optimization is designed to ensure the acceptance rate of service requests.The simulation results show that the proposed multi-stage migration architecture is superior in improving the prediction accuracy,and the performance of the energy-bandwidth ratio is improved by about 12%while ensuring a high service request acceptance rate.To sum up,this thesis studies the energy saving problem of network functional virtualization architecture based on resource prediction,and verifies the effectiveness of the migration algorithm and resource prediction model proposed in this thesis in reducing energy consumption and improving network service capability through simulation,which has certain significance and value for online migration research in NFV architecture.