Task Offloading Strategy In A MEC System With A Two-tier Edge Structure And Its Performance Study

With the rapid development for the technology of Mobile Edge Computing(MEC),MEC application scenarios have been more and more diverse and complex,among which the problem of task offloading is getting increasing attention from academia and industry.In order to improve the service experience of Internet of Things(Io T)users,ensure the stability of the MEC system and meet the performance requirements of tasks,task offloading strategy in a MEC system with a two-tier edge structure and its performance are studied.Firstly,for the MEC application scenario of the single-access Io T home with homogeneous-task,in order to ensure the uniformity and stability of the MEC system,by integrating edge-local vertical collaboration and edge-edge horizontal collaboration,a homogeneous-task offloading strategy in a MEC system with a two-tier edge structure is proposed.According to this strategy,a system model composed of a local computation model and an edge computation model is established to capture the workflow of homogeneous-task in a MEC system with a two-tier edge structure.Secondly,for the MEC application scenario of the hybrid-access Io T home,in order to meet the differentiated performance requirements of latency-sensitive tasks and latencytolerant tasks,a heterogeneous-task offloading strategy in a MEC system with a two-tier edge structure is proposed.By analyzing the workflow of latency-sensitive tasks and latency-tolerant tasks in the MEC system,a local computation model and an edge computation model are established for latency-sensitive tasks and latency-tolerant tasks,respectively.Then,by applying the Quasi-Birth-Death process and matrix-geometric solution method,the computation models are analyzed at steady state.The performance measures for evaluating the proposed task offloading strategies are derived,including the average delay of tasks,the utility of MBS cluster I and the average power of the MEC system.The numerical experiments are carried out to reveal the effects of different strategy parameters on the long-term performance of the MEC system.The simulation experiments are conducted to verify the rationality of the computation models.Comparing to traditional task offloading strategies and service policies,the validity of the proposed task offloading strategies is verified.Finally,an optimization objective function is given by balancing the average delay of tasks and the average power of the MEC system.Transforming the traffic load within the edge MBS layer and the average delay of latency-sensitive tasks into the equilibrium constraint and delay constraint with an adjustable upper bound,respectively,considering the steady-state constraints of the MEC system,the single-objective optimization problem with inequality constraints is formulated.The Powell-Hestenes-Rockafellar(PHR)algorithm based on the Lagrangian multiplier method is improved to give the optimization schemes of the proposed task offloading strategies.