Research On Mobility Support And Resource Allocation In Mobile Edge Computing Scenarios

In recent years,cloud computing has increasingly been pushed to the edge of the network,and the introduction of mobile edge computing has made it possible for many new applications to fall to the ground.However,the existing mobile edge computing architecture has problems such as excessive computation offloading delay and weak peak carrying capacity due to geographical dispersion and limitation of edge computing resources.Based on previous studies,this thesis starts with the "load prediction based on user movement" and "joint resource allocation and task scheduling",and proposes solutions to the above problems.The main work includes the following aspects:(1)Firstly,according to the dynamics of the edge network,the multi-level distribution of computing resources,the limited resource of a single server,and the minimum computing requirements of the service,it is proposed that edge computation offloading frameworks should be proactive and centrally controlled,and have the ability to make multi-level decisions,and think edge servers like a whole.Taking this as the goal,this thesis proposes a Multi-level Proactive MEC Offloading System(MPMOS),and introduces the main modules,system control flow,and user computation offloading progress of the system.The system works in an edge network with two layers of edge computing nodes,and controls the edge servers in the entire edge domain in a centralized control manner.The controller periodically predicts the workload of the entire edge domain.When demand changes,controller dynamically adjusts the computing resources,and designs the task scheduling scheme,which reduces the user's computation offloading delay and improves the ability of the edge network to carry the peak traffic.(2)The characteristics of the load changes of the edge network users,and the existing load forecasting schemes were analyzed.It is considered that the simple use of the historical statistical-based forecasting algorithm has a large limitation in the edge scenario.Under the MPMOS architecture,the existing research results of workload prediction and user mobility management are combined,and a Network Context based Workload Prediction algorithm(NCWP)is designed.The method predicts the user trajectory by using the user's network context information in a periodic manner to achieve the purpose of predicting the number of base station users per cycle.The map of Qingshuihe Campus of University of Electronic Science and Technology is selected as the simulation scene.Simulate user trajectories with SUMO software.The simulation results show that the NCWP can achieve higher prediction accuracy than the traditional ARIMA algorithm based on historical statistics.(3)This work analyzes the problems faced by computing resource allocation and task scheduling in the edge computing scenario,and summarizes the existing work.Under the MPMOS architecture,the delay of edge computation offloading is modeled by queuing theory.Solving the model requires mixed integer nonlinear programming,which belongs to NP-hard problem.In this thesis,the algorithm Joint Resource Allocation & Task Scheduling(JRATS)is proposed to solve the problem by solving the molecular problem separately,which based on greedy and optimization ideas.The MATLAB simulation method is used to evaluate the effect of the JRATS algorithm.Compared with the mode without cooperation between edge servers,it can improve the edge load capacity of the peak load,reduce user delay,and effectively integrate the resources of the edge.