Research On Task Offloading Strategy In Mobile Edge Computing

With the rapid development of Internet of Things and Mobile Internet,mobile communication network is facing the challenges,e.g.ultra-high connection density,ultra-low latency and higher user experience rate.Owing to the characteristics of low latency and network context awareness,Mobile Edge Computing(MEC)has become one of the key technologies for 5G networks to address these issues.To be specific,by transferring cloud computing services to the mobile access network edge,MEC enables traditional mobile networks to hold communication capabilities as well as IT services such as computing and storage capabilities.In this case,user equipment(UE)can offload computationally intensive or time-sensitive tasks to MEC servers with ultra-low transmission latency.In addition,MEC servers allocate computing resources as well as assist in task computation to speed up UE task processing and save energy consumption.Therefore,the MEC task offloading strategy has become one of the key factors determining the efficiency of the MEC system.By investigating the research status and characteristics of task offloading strategy in MEC and combining with the experience of successfully solving decision-making problems by deep reinforcement learning(DRL)in recent years,this thesis explores how to utilize DRL method to achieve optimal task offloading strategy within single-user and multi-user scenarios.The main contribution is presented as follows.(1)First of all,a task offloading strategy for single-user scenario is proposed.Jointly considering the scheduling order and migration decision of tasks,an energy consumption model with system,tasks and delays is constructed in the single-user MEC environment with multiple independent tasks.Besides,the task offloading strategy based on policy gradient reinforcement learning is designed to minimize the objective for task processing delay and energy consumption of UE.(2)Moreover,a task offloading strategy for multi-user scenarios is proposed.In the MEC environment of limited computing power and intensive user equipment,the allocation of computing resources in MEC server is considered as well as a Markov decision-making process is constructed.Consequently,a task offloading strategy based on DRL is designed to ensure the performance of delay-sensitive tasks in MEC system.(3)Finally,to verify the performance of both above strategies,the simulation of single-user and multi-user scenarios is built in Python environment.Additionally,simulation experiments are carried out respectively.As a result,the experimental results indicate that,compared with traditional methods,this strategy can significantly reduce the processing delay as well as energy consumption in single-user scenario.In multi-user scenarios,this strategy can effectively reduce the processing delay and timeout period of delay-sensitive tasks.In conclusion,by employing DRL method to design task offloading strategies in single-user and multi-user scenarios,both strategies have powerful adaptability to dynamic load changes.Furthermore,both are capable of effectively optimizing energy consumption and delay in task offloading process.Moreover,both strategies provide a potential to computing offloading service for future 5G network deployment scenarios such as Vehicular Networks,Internet of Things etc.