Research On 5G VNF Auto Scaling Based On Load Prediction

With the explosive growth of mobile applications,the frequency of people using phones is gradually increasing,and the demand for high bandwidth and low latency of wireless networks is becoming stronger.In order to meet user demand,domestic and foreign communication operators have accelerated the construction of 5G networks.How to effectively utilize existing equipment and save hardware costs in the process of network renewal is the main challenge faced by operators.5G networks based on NFV architecture can effectively deal with above challenge.The NFV architecture can standardize network functions,free them from specialized hardware and software,separate them from underlying devices,and allow for orchestration and management.NFV research includes the construction of 5G emulation platforms and the application of auto scaling technology.5G emulation platforms enable network testing and orchestration,and help operators achieve the automation and intelligence of network management.Foreign scholars have developed simulation software and MANO platforms such as Free5 GC,OSM,but these are independent and can’t provide a unified orchestration and management of network functions.The auto scaling technology can be used for network functions expansion and contraction,and balances the service volume and network capacity,but the widely used responsive auto scaling strategy has weak strain capability for unexpected situations and function singleness.To resolve above problems,this thesis proposes a VNF auto scaling system based on load prediction.The main research contents include:(1)Design a TCN-Attention load prediction model.This model combines the parallel computing advantage of TCN model with the attention mechanism’s ability to extract effective information.This model can accurately predict future base stations load based on historical data,and provide reference for base station VNF expansion and contraction.Compared with benchmark models,the TCN-Attention model has shorter training time and higher prediction accuracy.(2)Design a VNF scaling algorithm,and combine it with TCN-Attention model to form a predictive auto scaling strategy.The strategy can realize VNF expansion and contraction in advance,and solve the problem of insufficient ability with responsive auto scaling strategy.(3)Integrate OSM platform with 5G simulation software to establish a simulation platform from access to core network under NFV architecture.It enables the launch and maintenance of network functions,management and control of network traffic,access and connection of user devices,monitoring and collection of platform load.In addition,platform performance metrics such as up and down bandwidth were tested.(4)Implement a VNF auto scaling system based on load prediction.This system mainly includes load monitor,load prediction and auto scaling modules.It integrates auto scaling,5G network emulation and NFV architecture to realize the dynamic expansion and contraction of base station VNF.Compared with the auto scaling function of OSM,it has higher response accuracy and bandwidth utilization.