Research On Key Technologies Of Task Scheduling For Mobile Edge Computing

With the development of the Internet of Everything(Io E),the current network bandwidth can no longer meet the explosive growth of data.Moreover,high-tech technologies such as industrial Internet of Things,smart city and unmanned driving have higher demand for low latency.MEC deploys storage,computing,and network resources at the edge of the network,which is closer to users geographically.Offloading computation tasks to edge servers can enable tasks to be processed in time,which can effectively reduce end-to-end delay,improve network efficiency,and reduce the load on the cloud center.Computation offloading is the research focus in MEC.Designing an optimal offloading strategy in a dynamic environment can effectively reduce delay and energy consumption.At the same time,mobility management is also one of the difficulties in MEC.When users move between multiple edge servers,when and where to migrate services may affect the Qo E.This dissertation investigates the task offloading and service migration in different scenarios.For different MEC scenarios,the optimal offloading scheduling strategy is designed to reduce the delay and energy consumption of edge devices.On the other hand,with the constraints of migration cost,the optimal task migration algorithm is used to minimize the average delay.The contributions of this dissertation are as follows:First of all,for the coarse-grained offloading scenario,considering the time vary-ing channel and the interference between multiple users in a dynamic environment,this dissertation investigates the multi-user coarse-grained offloading problem.Users are regarded as players of the game,and the processing method(local or offloaded to edge server)is regarded as the strategy space of the game.The MEC-based multi-user computation offloading problem is modeled as an evolutionary game model to minimize delay and energy consumption.Then,through RD,the update process of the multi-user offloading strategy is studied,and it is proved that there is a unique ESS under the RD model.Finally,in the actual application environment,an evolutionary game theory based on Q learning(EGT-QL)is proposed.Each user independently selects and updates strategies based on Q learning.After continuous learning,it finally reaches ESS.Experiments prove the convergence of the proposed algorithm,and compare it with five related algorithms to verify the reliability of EGT-QL.Then,we study the fine-grained offloading problem in multi-user and multiserver scenarios,because it is not optimal to offload the whole task to edge server.This dissertation proposes a multi-user fine-grained task offloading scheduling method for mobile edge computing.In the joint offloading scheduling strategy,we model the computation task as a directed acyclic graph(DAG).Optimize the execution position and scheduling sequence of task nodes by analyzing the parallel processing of local and edge servers.Considering the energy consumption and delay,the computation offloading is regarded as a constrained multi-object optimization problem(CMOP),and then an improved fast non-dominated sorting algorithm with elite strategy(NSGA-?)is proposed to solve the CMOP.The proposed algorithm can realize local and edge parallel processing to reduce delay and energy consumption.Finally,a large number of experiments are used to prove the performance of the algorithm.The experimental results show that the algorithm can make the best decision in practical applications.Finally,we study the mobility management in MEC.When users are in the mobile state,computation tasks need to be dynamically migrated between multiple edge servers to maintain service continuity.Due to the uncertainty of movement,frequent migration will increase costs and delay,and non-migration will cause service interruption,so it is very challenging to design an effective migration strategy.This dissertation studies the multi-user task migration problem in a dynamic environment.Considering the migration cost,Qo S,migration load on the server,and spectrum resource allocation,a multi-user task migration model is proposed.With the constraints of migration cost,we describe the multi-user task migration problem as a minimization optimization problem to minimize system delay.In MADRL,this dissertation constructs an adaptive weight deep deterministic policy gradient(AWDDPG)algorithm to optimize the costs and delay.Using centralized training and distributed execution to promote collaborative communication between mobile users,through offline training of the AWDDPG model,mobile users can make migration decisions in real time.A large number of experiments show that our proposed algorithm greatly reduces service delay and migration cost compared with related algorithms.