Design And Implementation Of Service Migration Stratege In Mobile Edge Computing

In Mobile Edge Computing(MEC),service allocation and service migration are two important methods of service scheduling.MEC servers are located in close proximity to users,enabling users to seamlessly access services running on edge facilities.Constrains of storage and computing resources and user distribution inhomogeneity limit the number of services MEC servers can host at the same time.Therefore,these constrains result in high network delay as well as load balancing problem among servers.Moreover,when services migrate with user,there is a significant problem of migration costs because of the blindness of whether to migrate and migration target selection.Therefore,in MEC,service allocation and service migration are of great research significance.This paper summarizes and analyzes the relevant theories and research status of service allocation and service migration in MEC.To solve high network delay and load balancing problem,this paper proposes a service allocation algorithm based on cluster partitioning mechanism.In this algorithm,the whole area is divided into multiple clusters according to the service request quantity and location distribution of each sub-region.Then,the service initialization allocation based on storage and bandwidth constraint is carried out through the k-medoids clustering method.Finally,the initial allocation is dynamically adjusted by calculating the difference of service popularity and load information in each sub-region.Compared with some traditional heuristic algorithms,our approach can reduce the variance of the load on MEC servers by about 20%.Furthermore,in order to optimize the blindness of whether to migrate and migration target selection,this paper proposes a service migration decision method based on reinforcement learning.First,service migration is considered as a markov decision process.Based on this,the reward function is defined by communication cost,migration cost,user movement direction and the available resources.Finally,the migration decision is determined by simulating user mobile data training.The simulation results show that this method can reduce the migration cost.