Research On Adaptive Computation Offloading For Edge Networks With Heterogeneous Resources

Mobile edge computing(MEC)has shown its unique potential in serving new computation intensive tasks or delay sensitive tasks.Tasks from user equipment could be processed on edge equipment via computation offloading and service migration,thus breaking the limitations imposed by the limited resources and power of terminal devices.However,the heterogeneity of MEC systems,the dynamic nature of wireless environment,and the mobility of end-users all pose great challenges to the design of offloading and migration policies.This paper aims to reduce resource consumption of the edge network by studying computational offloading and service migration algorithms in heterogeneous networks.In the course of the study,the heterogeneity and density of the MECs are combined considered to make the offloading policy and migration policy more realizeable.This paper first introduces the research and development status of computation offloading and service migration and the basic principles of computation offloading and service migration strategy design.The relationships and processes of computation offloading and service migration are analyzed,as well as the basic techniques for implementing computation offloading and service migration.At the same time,the basic principles of computation offloading and service migration strategy design are introduced with reinforcement learning as the core,which lays the foundation for the subsequent research.Next,in this paper,two computation offloading methods are designed to takle the conmputation offloading problem for Edge Networks with Heterogeneous Resources.We first propose a greedy algorithm,in which each arrival task is greedily offloaded to the edge server with minimal utility,based on a global information of network states.While this greedy algorithm performs well in terms of system utility,the overhead incurred to collect the global information is not negligible.In addition to this,when the wireless environment changes drastically,the collected global information does not reflect the real situation of the network,thus causing a degradation of the algorithm's effectiveness.Inspired by this observation,we then propose a offloading algorithm based on reinforcement learning,which does not rely on such kind of information and can make offloading decisions based on learning experience.By so doing,the communication overhead can be largely reduced and the performance is also guaranteed.Finally,two service migration algorithms are also designed to address the service migration problem induced by user mobility and the relatively limited resources of edge devices during computation offloading.The first algorithm is still a greedy algorithm and achieves a good performance.However,like that in computation offloading,the overhead of this algorithm is a fatal problem.In view of this,we also design an algorithm for service migration based on the reinforcement learning approach.In contrast to computation offloading,we use the policy-based method of reinforcement learning,thus some of the problems that appear in the value-based method are averted and the algorithm achieves approximately 95%performance compared to greedy algorithms.