Research On Task Offloading Optimization Strategies For MEC With Dynamic Resource Allocation

Along with the rapid development of mobile communication technology since entering the21 st century,various computation-intensive applications,such as AR/VR,video stream processing,etc.,have been emerging continuously.The traditional model of mobile cloud computing results in large latency and energy consumption when dealing with such applications.To solve these problems,mobile edge computing(MEC)emerges as the times require.The MEC provides services for the mobile terminals by dropping the computing resources,disk resources and internet resources to the MEC servers.As the MEC servers are located at the edge of the network,it is closer to the mobile terminals.As a result,the data transfer distances are effectively reduced when processing computation-intensive tasks on such devices,that is,the transmission latency of computation tasks is reduced.At the same time,it provides resources such as CPU for mobile devices with limited computing resources and energy,which further minimizes the computational delay and energy consumption of local computing mode.In MEC,a highperformance task offload strategy can effectively reduce the task processing delay and energy consumption.Therefore,designing an efficient task offload strategy is the key to improve the performance of an MEC system.This paper focuses on how to design efficient offloading strategies for an MEC system in order to improve the efficiency of the MEC system.This paper designs high performance offloading strategies for different scenarios in MEC.(1)DRL-Based Computing Offloading and Resource Allocation PolicyIn this part,we design a computing offloading strategy based on the partial observable Markov decision process for multi-terminal devices and single MEC server.In this scenario,each terminal,as an agent,selects its own task offloading policy independently.We consider not only the influence of the task offload proportion on the performance of the task offload model,but also the influence of the information sharing model among end-devices on the performance of an offload policy.Finally,the problem is modeled as an A2C-based one named DAR-AC.Simulation results show that DAR-AC is correct and effective.Furthermore,compared with other proposed offloading policies,the results show that DAR-AC can effectively reduce the computational latency and energy consumption of a computing task.In addition,DAR-AC can improve its performance by freely exploring offloading strategies as the environment changes.(2)Computing Offloading and Resource Allocation Policy Based on Stochastic GameAiming at the resource competition among multi-terminal devices,a computing offloading model based on stochastic game is designed.In this part,the effect of resource allocation policy on the performance of an offloading policy in MEC is studied.Finally,based on the MARL in stochastic game,an algorithm named SGAR is designed to solve the issue.Simulation results demonstrate the effectiveness of SGAR and compare it with other two benchmark algorithms.Simulation results demonstrate the effectiveness of SGAR.Compared to only-edge and onlylocal,SGAR has lower computational latency and energy consumption.Moreover,it can achieve Nash equilibrium in limited steps,which improves the performance of the whole MEC system.