Computing Offloading And Service Placement In Mobile Edge Computing

Mobile edge computing will ”sink” services located in the cloud data center to the edge of the mobile network,providing computing,storage,network and communication resources at the edge.MEC emphasizes proximity to mobile users,so as to reduce the delay of network operation and service delivery and improve the user service experience.Delay sensitive artificial intelligence(AI)applications,such emotion recognition and face detection,are usually limited by energy consumption and computation resources,and cannot meet the Qo S requirements.MEC aims to solve this problem.Many researchers have put forward solutions of mobile edge computing offload and migration.However,the existing mobile edge computing offload and migration schemes have the following problems: 1)most of the existing service migration strategies are based on user mobility,without considering the personalized and elastic allocation of the computing services;2)the limited computation and storage resources of the edge nodes and mobile devices make the deployment of AI applications at edge is difficult to achieve,so we should consider the segmentation of AI tasks and the fine-grained collaborative service placement;3)the computing offload problem of emotion analysis service;4)the failure of computing task offload due to user mobility.In the face of these problems,this thesis studies from four aspects:1.This thesis studies the dynamic service migration of edge cognitive computing.This thesis proposes an edge cognitive computing architecture,which deploys cognitive computing on the edge of network to provide dynamic and elastic storage and computing services.Considering the elastic allocation of computing services and the mobility of users,this thesis proposes a service migration learning strategy based on Q-learning.The experimental results show that compared with non migration and random migration,the migration strategy proposed in this thesis has greater performance benefits,can significantly reduce the overall service cost,and can adapt to the change of mobile access mode,improve the quality of service.2.This thesis studies the collaborative service placement and migration of edge clouds.Considering how to deploy computing intensive deep learning model in edge devices with limited resources,this thesis divides the complex AI model into smaller subtasks,and cooperatively executes them among edge nodes,and proposes a cooperative service placement strategy for efficiently deploying various subtasks among edge nodes.In order to overcome the challenge of lack of system side information,we describe the problem as a combined context multi arm bandit(MAB)problem.On this basis,an online learning algorithm based on Thompson sampling and greedy method is proposed.Theoretical analysis and evaluation show that the algorithm can effectively learn and adapt to the dynamic network environment and user demands,while meeting the minimum service needs,reducing the delay and migration costs.3.In this thesis,the multi-level computing offload problem of emotion analysis service is studied.Considering that at different levels,latency,energy consumption and quality of service(such as accuracy of emotion recognition)are different.In this thesis,we construct a delay constrained energy minimization problem to implement task scheduling strategy among multiple layers.The experimental results show that,compared with local execution strategy,cloud execution strategy and random execution strategy,multi-level task offloading strategy can always find a better energy-saving scheme.The emotion analysis application under this architecture can meet the requirements of user delay and emotional service quality while reducing energy consumption.4.This thesis focuses on the research of mobile ad hoc cloudlet computing offloading based on opportunism.Considering the unreliability of edge computing based on D2 D,this thesis proposes a mobile ad hoc cloudlet computing offload mode based on opportunism,and presents a delay analysis and computing offload mode selection algorithm based on the user mobility model.The simulation of OPNET verifies the effectiveness and practicability of this scheme to improve the service quality of mobile users.Compared with other computing offload modes,this mode can support users' high mobility and guarantee the demand of task delay,and has better flexibility and higher performance.To sum up,the mobile edge computing offload and migration strategies proposed in this thesis make full use of the characteristics of user mobility and cognitive computing services,develop dynamic service placement and offload strategies for AI application services,provide elastic computing resources,and improve the utilization of edge cloud computing resources and user Qo S.