Research On Computation Offloading Mechanism In Edge Networks

With the exponential growth of IoT devices and the continuous emergence of new mobile applications,the contradictions between the limited resources of individual termi-nal devices and complex application requirements have become prominent.To address the issue,computation offloading was proposed,which enables terminal devices to trans-mit application-related data to other computing entities,such that utilize other computing resources to execute its applications.This provides the possibility to expand the comput-ing capabilities and extend the battery life of terminal devices.In computation offloading,the optimal decision is required to be made for some key issues,such as whether the tar-get terminal device performs computation offloading,when to computation offloading,and where to computation offloading,so as to truly realize the expectations of edge com-puting.However,different system metrics,network environment and available resources will have significant impacts on the offloading decision.Therefore,it is of great signif-icance to explore the mechanism design of computing offloading under different system configurations.On the other hand,edge computing,as a new computing architecture,allows the of-floading to be done within the edge networks,thereby reducing the communication time between the offloading device and the other computing entities.Therefore,this disserta-tion aims to reduce the energy consumption of terminal equipment and ensure application service requirements under different scenarios,based on convex optimization,Markov decision process,and reinforcement learning theories.By means of model construction,algorithm design and optimization,theoretical analysis and simulation verification,the research on computational offloading mechanism is carried out.The main contributions of the dissertation include:1)For the edge computing offloading with limited frequency channel resources,a joint multi-user offloading decision and wireless time-frequency resource allocation scheme is proposed,such that minimize the total energy consumption of the users and achieve good fairness fairness of completion time among different applications.By decomposition,the optimal offloading decision and channel allocation are derived when considering orthogonal channel allocation.Considering the limited channel re-sources,we propose that same frequency channels can be time-shared by multiple devices.Using branch and bound and successive convex approximation methods,the approximate optimal solution on multi-user offloading decision and time-frequency resource allocation is obtained,such that minimize the total energy consumption of the users.In order to reduce the complexity of problem solving and further satisfy the fairness of the application completion time,the channel allocation is decoupled from the offloading decision and time allocation,and a two-layer heuristic algorithm is proposed.The simulation results verify the effectiveness of the proposed algorithm in terms of reducing system energy consumption and achieving fairness of application completion time.2)For the peer-to-peer computing offloading without central control,two distributed mechanism on application partition and collaborative device competition is proposed to solve the problem of computing offloading caused by incomplete information in a decentralized network.Various factors,such as the application deadlines,the comput-ing capability constraint and the energy limit of each collaborative device are com-prehensively considered to formulate the optimization problem of joint application partition and collaborative device selection.Based on the paired energy consumption of the offloading device and the collaborative device,and the system-level energy con-sumption,respectively,two distributed application partition and collaborative device selection algorithms are devised.The performance lower bound of the two algorithms and the performance gap compared with the optimal solution are theoretically proved,and the computational complexity is further analyzed.The simulation results reveal the trade-off between system-level energy consumption performance and algorithm complexity.3)For the peer-to-peer computing offloading with unknown future system information,a dynamic local computing frequency and offloading transmission power strategy is proposed to meet the challenge of computing offloading caused by stochastic property in wireless networks and computing resources.Considering the randomness of con-tinuous task arrival,the dynamic availability of nearby idle peer devices,and the non-stationary wireless interference between offloading devices,a learning-based method is used to make each offloading device dynamically adjusts the local computing fre-quency and transmission power strategy based on local observation,in order to learn the environmental information,such as wireless network conditions and computing resource availability.By introducing post-decision states,the partially known infor-mation of the system is characterized,which accelerates the learning process.The simulation results verify the effectiveness and scalability of the proposed algorithm.4)For the end-edge cooperative computing with system statistics information,the op-timal end-edge computation offloading strategy is proposed to converge the hetero-geneous resources in the edge network.A novel cooperative computing paradigm is proposed.According to the different states of peer collaborator,it can be utilized to perform communication assistance and computing assistance,so as to make full use of heterogeneous communication resources and computing resources.Based on the stochastic prior information of wireless channel state and the availability state of peer devices,the computing offloading model based on Markov decision process is for-mulated.the optimal end-edge computing offloading strategy is theoretically derived.The impact of different system parameters on the offloading decisions is quantified.The simulation results verify the performance gain brought by the end-edge cooper-ative computing,and show that the proposed mechanism can effectively reduce the energy consumption of the terminal devices while meeting the application deadline.