Research On VNF Demand Forecast And Migration Method Of Edge Network

In the era of the Internet of Everything,with the emergence of a large number of smart terminals and various real-time network services,the current bloated network is not only difficult to support the needs of technologies such as Virtual Reality(VR)and smart medical care,but also in terms of network maintenance upgrades and expansion.Network Function Virtualization(NFV)technology emerged at the historic moment,aiming to implement various network functions running on proprietary hardware through software,which not only improves the scalability of the network architecture,but also makes the network more It is flexible and reduces system costs.It can use the Edge Computing(EC)paradigm to sink real-time network services to the edge of the network,thereby improving the quality of user service experience and reducing user-side costs.This thesis focuses on the prediction and migration methods of Virtual Network Function(VNF)in the edge network.Its important work is as follows:First,examples of various virtual network functions placed at the edge of the network can provide end users with different network services.Deploying an appropriate number of VNFs in advance can not only reduce system costs,but also improve network service quality.Aiming at the problem that the current VNF demand forecasting method is low in accuracy and not suitable for edge networks,this thesis proposes a VNF based on the combination of Support Vector Regression(SVR)and Gated Rrecurrent Unit(GRU)models for edge network scenarios.Demand forecasting methods.This thesis considers that network edge traffic has the characteristics of burstiness,self-similarity,and long correlation.Combining the advantages of the SVR and GRU models,the SVR and GRU models with lower computational complexity are used to extract the historical time series data of network services.Short-term features and long-term features to improve the accuracy of VNF demand forecasts and realize the advance placement of VNFs in the edge network.The simulation results show that the prediction method proposed in this paper can reduce the error of 15%-18% compared with the Long Short Term Memory(LSTM)network model,and has better prediction effect.Second,when considering the overload of edge server nodes in the network,it cannot meet the needs of all users under its coverage,Therefore,it is necessary to migrate or copy VNF to ensure network service quality and realize load balancing.This thesis weighs the cost of VNF replication and the degradation of network service quality caused by VNF migration,and cooperates with VNF replication,and proposes a VNF migration method based on deep reinforcement learning.This method first selects suitable target server nodes through threshold analysis based on idle resources and load,and constructs a Markov decision process based on the optimization of delay and cost,and finally through Deep Q Learning(Deep Q-learning)realizes the selection of the optimal strategy.The simulation results show that,compared with the traditional VNF migration method,the VNF migration method proposed in this thesis can achieve server load balancing at a lower cost under the premise of ensuring the delay requirement of the service function chain.