Research On Microservice Wireless Network Architecture And Reliability Mechanism

The increasingly rich mobile network application scenarios put forward higher requirements for network flexibility,agility and scalability.Network virtualization and cloud have become an inevitable trend.To flexibly respond to the demand for more diversified functions,quality of service,management strategy,deployment mode and capability opening in future applications and scenarios,service-based radio access network is becoming the research focus of future network architecture.However,at present,the architecture design,function definition,interaction mode and management mode of service-based RAN have not formed a unified standard.The resource fluctuation of cloud environment also brings great challenges to the guarantee of quality of service(QoS)and reliability of service-based RAN.Therefore,the main research content of this thesis is the reasonable design of service-based RAN architecture and the reliability guarantee mechanism of service-based RAN QoS during its cloud native deployment.To solve the problem of service-based RAN architecture design,this thesis proposes a Microservice Based Radio Access Network architecture(SBA-RAN).Firstly,SBA-RAN realized its service definition according to the service capability of each layer of the existing protocol stack.Secondly,SBA-RAN introduced a unified standardized interface into each RAN micro-service,which met the functional interaction requirements between micro-services.On top of that,SBA-RAN implemented microservice management and user-plane traffic scheduling by introducing microservice instance management entity.In addition,to verify the advantages of SBA-RAN compared with C-RAN architecture in terms of network energy efficiency,this thesis designed an elastic scaling strategy of SBA-RAN micro-service based on deep reinforcement learning(DRL)which can adjust the computing resource allocation of micro-service instances in real time according to the user requirements.Simulation results show that compared with C-RAN architecture,SBA-RAN has improved network flexibility and energy efficiency,and the strategy proposed in this thesis can improve network energy efficiency on the premise of guaranteeing network throughput.Secondly,to solve the QoS reliability guarantee of service-based RAN during its cloud native deployment,this thesis proposes a DRL based cloud task scheduling strategy in the public cloud scenario.In the public cloud platform,multiple services coexist and are highly dynamic.This strategy comprehensively considers the resource-service quality model of different cloud services.A resource competition model of multiple services coexisted in the public cloud scenario was established,the QoS optimization problem of service-based RAN was modeled as a Markov decision process of parameterized action space,and a task scheduling strategy was designed to ensure the QoS of service-based RAN.This strategy can be adapted to more general cloud-network integration scenarios.Compared with other DRL algorithms,this strategy adopts multi-pass mechanism at the network input layer to jointly optimize discrete and continuous actions,eliminate the influence of irrelevant parameters on the Q value of discrete actions,and has better convergence.On the premise of ensuring the normal operation of other services in the cloud platform,the QoS reliability of service-based RAN is greatly improved.