Study On Dynamic Offloading Strategy In Mobile Edge Computing

Mobile Edge Computing(MEC),as a supplement to mobile cloud computing,is one of the key technologies in 5G systems.It aims to distribute some resources of the remote center server to edge servers deployed on the wireless access network and closer to the device,reducing the delay caused by long-distance transmission of tasks,saving channel resources,reducing energy consumption of the device,and also bringing a better experience to users.Among them,computing offloading is a key technology in MEC.By offloading tasks in mobile devices to the MEC server for processing,it greatly solves the problems of high latency and low energy efficiency caused by insufficient computing power of mobile terminals.However,when the computing requirements of users is increasing,MEC server resources and channel resources are limited,and energy consumption and latency time are wasted during the offload transmission process.For these reasons,how to design a reasonable computing offloading strategy for the MEC system to meet user needs is the key point of this question.Previous studies often only considered that mobile devices only processed a single computing task at the same time.In reality,there are often multiple parallel computing tasks that need to be processed by the device at the same time,which puts a test on the stability of the device system.Therefore,the main content of the thesis is how to make offloading decisions and allocate resources reasonably for the mobile device to achieve low latency,high energy efficiency and stability when mobile devices are running multiple applications.First,from the perspective of high energy efficiency of mobile devices,the thesis considers how to take into account the stability of mobile device task buffer queues.Next,the thesis further considers mobile device energy consumption and latency as optimization goals in the MEC system with the support of energy harvesting technology.The target is how can make the mobile device allocate resources reasonably and simultaneously satisfy the stability of the battery energy queue and the buffer queue.The research content and results of the thesis are described as follows:(1)The thesis proposes an online offloading MEC model.First,in this system,the mobile device makes offloading decisions to minimize the power of the device by evaluating the system status in this time slot such as the channel status,backlog buffer queue status and the amount of receiving tasks.In the thesis,the optimization goal is to minimize the long-term power of the device,and the stability of the task buffer queue is used as a constraint to propose an optimization problem.Then we transfer the problem into a new version based on Lyapunov optimization theory while the Dynamic Minimization of Power(DPM)algorithm is proposed.The solution gap between the new problem and the original optimization problem has been analyzed,and so has been the upper bound of the buffer queue length.In the simulation part,the thesis first analyzes the gap between the proposed algorithm solution and the original problem solution,while the analysis of the proposed weight on power and buffer queue length has been done also.In the comparison,the results show that the DPM algorithm perform better than other traditional offload algorithms in the energy efficiency.(2)In the thesis,we propose an efficient tasks allocation strategy in an MEC system with energy harvesting to minimize the weighted sum of energy consumption and computation delay for mobile devices.In addition,we design a queue for the coming tasks from which the devices fetch the tasks to execute.Based on the Lyapunov optimization method,we propose an online Dynamic Lyapunov Optimization-based Tasks Allocation(DLOTA)algorithm,to determine the the tasks allocation strategy by adjusting both the CPU frequency and the offloading transmission power of mobile devices.The advantage of DLOTA algorithm is that the tasks allocation decision only depends on the current system state and does not need to predict the future state.Simulation results show that the proposed model and algorithm can achieve the stability of battery energy levels and the trade off between energy consumption and execution delay.