Research On Controller Placement And Edge Selection In Multi-access Edge Computing Based On Software Defined Networks

With the popularization of smart mobile terminals such as smart phones or tablets,the Internet of Things(Io T)has developed rapidly.Traditional cloud computing technologies have been unable to meet the requirements of "low-latency,largebandwidth" cloud resources on the edge side.The multi-access edge computing is proposed by European Telecommunications Standards Institute(ETSI).It can provide IT service environment and computing capabilities at the edge of mobile networks to reduce network operation and service delivery delays.Software defined networking(SDN)technology is applied to multi-access edge computing.It can support the access and flexible expansion of massive network devices and provide efficient.The efficient and low-cost automated operation and maintenance management is provided.SDN can improve the network performance of multi-access edge computing in complex edge network environments.In the multi-access edge computing environment based on SDN,how to effectively and reasonably deploy SDN controllers and improve network performance has become an urgent problem.In the multi-access edge computing based on SDN,there are multiple ways for tasks to access the edge computing environment.The task stability is improved and the system energy consumption is reduced by rationally selecting edge nodes for task execution.Therefore,it is of great theoretical and practical significance to study the controller placement and edge selection under the multi-access edge computing based on SDN.In order to solve the above problems,the main research contents of this thesis include the following aspects:(1)The network performance is improved by properly setting the number of SDN controllers,controller locations and the mapping relationship between switches and controllers.The SDN controller placement algorithm based on deep reinforcement learning is designed in this thesis.Firstly,the deep learning technology is used to predict the network traffic load.The controller placement model is established to minimize the average network delay and improve the controller load balancing effect.Secondly,the state space,the actions of controller placement and reward functions are designed based on the network topology and the objective function.The SDN controller placement scheme is obtained through deep reinforcement learning technology.The network performance of the multi-access edge computing environment based on SDN is improved.Finally,the proposed algorithm is experimentally verified.In the experiment of controller placement algorithm based on deep reinforcement learning,the controller placement algorithm in this thesis is compared with k-means,particle swarm optimization(PSO)and the simulated annealing partition-based k-means(SAPKM).The algorithm proposed in this thesis is better than the comparison algorithms in terms of average latency,control plane utilization and load balancing.(2)In order to improve the stability of mobile device tasks and reduce the total energy consumption of the task execution,an efficient edge selection algorithm based on task stability and energy consumption minimization is proposed in this thesis.Firstly,the information about mobile devices and edge clouds is collected by SDN controller.The channel estimation is performed on the wireless communication network,and a channel estimation matrix is calculated.Secondly,the task queue model,the local task execution model and the remote task execution model are established respectively.According to the above model,the objective function of the edge selection model is constructed.An edge selection solution based on Lyapunov's optimization is proposed to realize the multi-task allocation on multiple edge computing nodes.This solution can effectively improve task stability and reduce total energy consumption.Finally,the proposed algorithm is experimentally verified.In the edge selection experiment based on task stability and energy consumption minimization,the algorithm in this thesis is compared with the learning automata based decision making algorithm(LADMA),the energy-efficient task offloading algorithm(ETO)in a non-SDN environment,the latency aware task assignment algorithm(LATA)and the SDN based assisted offloading algorithm(OAOP)in an SDN environment.The experimental results show that the proposed algorithm in the thesis has a better performance in terms of task stability and energy consumption.The task stability is improved effectively,and the system energy consumption is reduced.