User Computing Offload Traffic Prediction Based On Network Function Virtualization In Mobile Wireless Networks

With the rapid development of 5G,three kinds of services will be deployed including Mobile Enhanced Broadband,Ultra-High Reliability Low Latency Communication and Massive Machine Communication.As a key technology for 5G,the convergence of Edge Computing and Network Function Virtualization enables users with more computing intensive services by deploying virtual network function components in mobile access networks.The convergence of Edge Computing and Network Function Virtualization provides users with computing intensive services in close proximity by deploying Virtual Network Function components in mobile access networks,which leads to complexity on network management and orchestration.Since computational capacity and storage capacity of MEC servers are usually limited,how to collaboratively utilize the limited MEC server resources in a multi-MEC server scenario is a challenge for Service Function Chain deployment and optimal VNF placement.User computing offload traffic is indispensable to solve the above problem.Focusing on the user computing offload traffic prediction,the main contributions of the thesis are listed as follows:(1)A systematic simulation platform for computational offloading in mobile radio access networks is constructed to provide the datasets for computational offloading traffic prediction,which is based on a multi-cells and multi-user scenario.Different user mobility is supported.Statistical analysis are presented on the datasets generated from systematic simulations.(2)A Dense-Sparse Sampling-Based Feature Extraction Algorithm For VNF Traffic Time Series is proposed,and the DSSFE algorithm is evaluated by using both the public temporal prediction dataset ETThl and the simulated VNF dataset,respectively.The evaluation results show that the proposed algorithm has optimal prediction results in terms of MSE compared with baseline algorithms including RNN and LSTM.(3)Considering that user service requests are processed collaboratively among base stations,a VNF traffic prediction algorithm based on Multi-path and Spatio-Temporal Graph Fusion is proposed,which extracts and fuses the temporal and spatial features under different paths effectively.The performance of MSTGF algorithm is verified in terms of MSE compared with the baseline algorithms.