Deployment And Optimization Of Control Plane In Edge Computing-based Software-defined Internet Of Vehicles

The Internet of Vehicles(Io V)is a heterogeneous network consisting of a large number of high-speed mobile nodes,requiring the management of various networking devices and addressing the data transmission needs of diverse applications.With the continuous expansion of the Io V,traditional network architecture-based solutions cannot effectively manage numerous mobile devices.Additionally,the uneven distribution of network traffic caused by high-speed movement and uneven distribution of vehicles hinders traditional vehicle networking from fully utilizing network resources to provide high-quality services.Software Defined Network(SDN)is a novel network architecture that separates the control plane from the data plane,enabling flexible and unified management of network devices through the control plane.It also guides network traffic based on global network decision-making,ensuring full utilization of network resources.In SDN,the control plane exhibits logical centralization,and the frequent changes in network topology leading to significant overhead between different controllers.The uneven distribution and rapidly changing flow state in the Io V results in considerable differences in the load of different controllers in the control plane,affecting the performance of the control plane and the efficient utilization of controller resources.To ensure the performance of the control plane and fully exploit the advantages of SDN,this thesis studies the deployment and optimization of control plane in Software-Defined Internet of Vehicles(SD-Io V),aiming to deploy appropriate control plane based on network status and continuously optimize control plane according to changes in network status.The main work of this thesis is as follows:(1)Aiming at the problem that different structural control planes adapt to different scenarios,this thesis analyzes application scenarios in intelligent transportation system(ITS).Based on the coexistence of multiple application requirements,this thesis designs a hierarchical structure for ITS.Subsequently,a three-layer control plane for SD-Io V is proposed.Considering the ultra-low latency requirements of the Io V and the advantages of edge computing in providing ultra-low latency services,this thesis proposes to deploy the control plane to the edge layer.To demonstrate the advantages of this structure,a comparison of state synchronization overhead for different control plane structures is conducted.Moreover,a distributed path calculation scheme for hierarchical control plane is proposed and validated.(2)Based on the proposed three-layer control plane,this thesis models the performance of the control plane from three aspects: delay,load difference,and control reliability,addressing the significant impact of vehicle mobility and flow changes on control plane performance.A controller deployment problem is constructed based on these three aspects.Considering the dynamic nature of the network state of the Io V,a dynamic controller deployment problem is constructed using Markov decision processes(MDP),and a dynamic controller deployment algorithm based on deep reinforcement learning(DRL)is proposed,so as to dynamically adjust the number of controllers and deployment locations according to the network status.(3)Aiming at the problem that the network status of different regions has a great influence on each other,and centralized deployment can easily lead to the problem of excessive load on the single-point server,this thesis proposes a distributed controller deployment algorithm based on multiagent deep reinforcement learning(MADRL).By modeling the controller placement problem as a Markov game process,it reduces the mutual influence between different network regions during the controller deployment process and ensures that intelligent agents are constrained by the overall performance of the control plane when making individual decisions,and further improve the efficiency of algorithm execution.Through this distributed algorithm,the problem of untimely collection of network information is also avoided,and the performance overhead of the server when deploying the controller is reduced.(4)To continuously optimize the control plane with less overhead,this thesis presents a load balancing algorithm for hierarchical control plane.By migrating network devices on a small scale and adjusting the dependency relationship between controllers and network devices based on network status,continuous optimization of control plane performance is achieved,and precise adjustment of high-load controllers is realized through multi-agent deep reinforcement learning.Figures 64,tables 23,references 148.