Research On Computation Offloading And Resource Contribution Strategies In Mobile Edge Computing

With the maturity of Internet of Things(IoT)and 5G cellular communication systems,a variety of latency-sensitive and computation-intensive applications(e.g.,virtual reality,augmented reality,video processing,and dynamic control services)have been deployed on the IoT devices.The transmission and processing of the data generated by IoT devices brings a significant challenge to network load and device battery life.It is generally difficult and even impossible for these IoT devices to process all the application tasks locally while satisfying the performance demands of applications.Therefore,the technology of mobile edge computing(MEC)is born.MEC deploys the intensive computation-resources close to the edge of wireless networks(e.g.,the cellular base stations or the local wireless access points),which has been envisioned as a promising paradigm to address the shortcoming of the conventional cloud systems and eventually enable the computation-intensive and latency-sensitive mobile Internet service.Meanwhile,the massive amount of data generated by IoT devices has driven the development of federated learning(FL),and FL combined with MEC has received a lot of attention.Task offloading and resource contribution for IoT devices have become hot issues for current research.In this paper,based on the summary and investigation of the current status of domestic and international research on task offloading and resource allocation in MEC.The following work is carried out:(1)In a cellular cell system with multiple IoT devices and multiple BSs,the IoT devices generate splittable offloading tasks that can be divided into subtasks and executed locally or offloaded to BSs for parallel processing.An offloading delay optimization method is proposed in this paper.We theoretically analyze the properties of the offloading delay optimization problem and obtain a satisfactory offloading strategy by designing a task offloading algorithm.A real dataset from the CBD area of Melbourne is used for evaluation.Compared with various offloading algorithms,the experimental results show that the DOLA algorithm can converge quickly and is adaptive when the system scale increases.(2)With the rapid development of MEC,the data generated by devices provides a important nudge to ML,and thus FL has gained great attention.In a cellular system that combines MEC and FL,the base station(BS)issues an FL task with a payment,and users compete for the payment by contributing computation resources and local data.In this paper,two utility functions are designed for users and BS respectively.Users and BS continuously adjust their strategies to obtain the maximum utility.The policy determination process of BS and users is formulated into a Stackelberg game.More importantly,the existence of Stackelberg equilibrium is proved theoretically.A algorithm named OIMA algorithm is designed to allow BS and users to obtain satisfactory policies in finite iterations.Simulation experiments show that the OIMA algorithm in this paper can obtain higher utility values compared with other incentive methods.