Task Offloading Of Multi-access Edge Computing In 5G Networks

With the rapid development of 5G and the Internet of things(IoT),the number of edge devices and data generated by these edge devices are growing explosively.Some new applications,such as AR,VR,Autonomous Vehicles(AVs),industrial Internet,telemedicine,smart home and so on,put forward new requirements for response delay and data security.In this context,edge computing has become a key technology to solve these problems,which has attracted more and more attention.In a wide area of edge computing application scenarios,such as Uavs,the edge layer is in the form of server cluster,rather than a single server node.When there are a large number of mobile devices,and at the same time,there are a lot of tasks need to be offloaded of these mobile devices,research on reasonable task offloading algorithm deployed on the mobile devices is still a hot topic.In the existing literature,classical task offloading algorithms such as static Round Robin algorithm,Min-Min algorithm,dynamic Least Connections algorithm,Ant Colony algorithm and so on,all realize shallow level load balancing.In view of this problem,this paper proposes a task offloading strategy JD-UCB1 based on enhanced learning UCB1 algorithm to achieve deep level load balancing.Based on the task offloading model we designed and the clear definition of the problem,we propose that the problem of task offloading in edge computing can be attributed to the problem of “exploration” and “exploitation” of servers.Therefore,referring to the UCB1 algorithm of Multi-Arm Bandit(MAB)problem,we design a dynamic adaptive task offloading strategy named JD-UCB1.Then the overall load balancing performance and real-time load balancing performance are verified by experiments,and through the comparison with Round Robin task offloading strategy,we prove that its performance is more superior in stability,adaptability and scalability.The experimental results show that the application of reinforcement learning algorithm to task offloading is a good direction to achieve deep load balancing.