Task Offloading Strategy Based On Task Classification And D2D Assistance In MEC System And Its Performance Study

As a new network architecture,multi-access edge computing(MEC)sinks the computing power and storage resource of remote cloud data center to the edge of mobile network,thereby shortening system response time,saving bandwidth resource and providing users with better quality of service(Qo S).Under the cloud-edge-end collaborative architecture,this thesis introduces device-to-device(D2D)communication technology,explores the offloading strategy of MEC system and studies its performance evaluation and optimization issue based on non-offloading and offloading tasks.Firstly,considering a small and simple resource edge service platform,in order to enhance user experience and achieve real-time and efficient task processing,a MEC task offloading strategy for homogeneous edge servers is proposed.Considering the burst arrival mode of tasks,we extend Poisson process to the Markov modulated Poisson process(MMPP),build a local model based on preemption priority for terminal device(TD)and D2 D device,and build an edge cloud model based on the continuous work of service desks for edge server.Secondly,in view of the diversity of network edge resource and the difference of computing power in complex MEC application scenario,a MEC task offloading strategy for heterogeneous edge servers is proposed.In order to save energy at the edge,set the edge server to switch between the ON and OFF states.Driven by MMPP,considering the endurance of task waiting,an impatience mechanism is introduced into TD,and a local model based on midway exit and preemption priority is constructed.Considering the energy conservation issue at the edge,an edge cloud model based on partial service desks shutdown is constructed.Thirdly,aiming to verify the effectiveness of the proposed MEC task offloading strategy,the matrix geometric solution method is used to analyze the queuing model and give the steady-state solution of the model.The mathematical expressions of performance metrics such as task average delay,TD average energy consumption and edge energy saving rate are derived.Based on the theoretical analysis result,the perfoemance evaluation experiment is conducted to study the trend of various performance metrics,and the impact of strategy parameters such as offloading probability and arrival rate on system performance is revealed.Finally,based on performance metrics such as delay,energy consumption and energy saving rate,and considering the different needs of network users,a nonlinear multiobjective optimization problem is proposed.The fast and elitist non-dominated sorting genetic algorithm(NSGA Ⅱ)is improved from aspects such as initialization,crossover and mutation.The effectiveness of the improved algorithm is verified by comparing the convergence of the fitness function.Based on the performance evaluation result,the optimization experiment is conducted to provide a set of Pareto optimal schemes of strategy parameters.We can determine the offloading scheme based on the preferences of user and service provider.