MEC Server Placement And Overall Task Offloading Based On Reinforcement Learning

With the rapid development of Mobile Internet and the rapid popularization of smart mobile devices,there has been an explosion of mobile communication traffic,and a large amount of data is expected to be generated at the edge of the network.As a result,traditional cloud computing technologies cannot meet the low-latency requirements of mobile users.To reduce the latency,Mobile Edge Computing has been proposed to offload part of the workload from mobile devices to nearby edge servers which have more powerful computing resources.As an essential component in the 5G architecture,MEC has always been a research hotspot in academia and industry.Focusing on the optimization of the deployment of edge servers in MEC scenarios,the paper designs a deployment scheme based on reinforcement learning theory.Firstly,we propose a MEC system model,which takes the calculation delay of base stations as optimization goals,and considers placing the server in LTE Macro Base Station,which cannot only balance the workload of the edge server but also reduce the latency of users as much as possible.Then,we formulate the edge server deployment problem as a single-objective optimization problem.We define actions,states,rewards according to the reinforcement learning framework,and propose the Q-ESP algorithm based on the Q-Learning algorithm.Finally,compared with other algorithms through simulation experiments,it is verified that the Q-ESP algorithm proposed in the paper has a better effect on reducing user delay and balancing edge server load.Aiming at the single-cell computing offloading problem in the MEC environment,the paper proposes an offloading scheme by analogy to the MAB problem.In the single-cell scenario,each offloading decision of the computing task is regarded as one arm in the MAB problem.By comparing the solving algorithm of the MAB problem,the paper innovatively proposes an offloading strategy called UTO that considers the amount of task computation.The algorithm determines the task to be selected in the next time slot based on the reward value of the task,the number of times the task is selected,and the amount of task computation.The simulation results show that by taking the task calculation as a determinant of the offloading strategy,the processing delay of tasks can be reduced effectively.