Research On Computation Offloading In Ultra-dense Edge Computing Networks

With the rapid development of technologies like 5G,artificial intelligence and Internet of Things,the number of intelligent terminals has shown explosive growth,which brings serious challenges to existing communication networks and cloud computing service models.In order to cope with stringent delay constraints,crowded network bandwidth and limited computing and energy resource on terminal equipments,the edge computing technology that offload partial or all of the computation-intensive work to the edge computing equipment rather than the remote cloud computing facility was proposed.However,as an emerging network architecture,there are still many open problems in fulfilling effective computation offloading in the edge computing networks.First,the communication and computing process need to be modelled jointly.Second,due to the dynamic computation tasks and the interplay among large amount of nodes in ultra-dense edge computing networks,the dynamic optimization of computation offloading is prohibitively complicated.Last but not the least,market price factors also have impact on computation offloading strategies.Most of the existing research on computation offloading is limited to ordinary static scenarios.The centralized optimization of a small number of users cannot fully and effectively solve these challenges.Therefore,this thesis explores the optimization of distributed dynamic computation offloading in ultra-dense edge computing networks in two scenarios,namely computation offloading from the network to edge computing nodes in downlink and computation offloading from smart mobile devices to edge computing networks in uplink,respectively.In the scenario of downlink computation offloading,a dynamic model for computation service price and available energy is designed first,and a utility function including both computation service reward and energy cost is proposed as the optimization target.The investigated distributed optimization problem for dynamic computation offloading is constructed as a multi-player dynamic stochastic game.Then,the mean field game method is introduced to effectively transform the N-player dynamic stochastic game into a two-body game,with the optimal solution to achieve Nashequilibrium.We further propose an iterative algorithm for solving the mean field equilibrium,and the optimal computation offloading strategy based on mean field game is obtained accordingly.Simulation results not only show that the proposed offloading strategy has the best cumulative utility,but also evaluate the important influence of the uncertainty of energy state evolution.In the scenario of computation offloading from smart mobile devices to edge computing networks,the dynamic wireless channel and the computation price are modeled first,and the transmission energy consumption and computation cost are combined as cost functions.By taking the dynamics of wireless channels and computation task queues as the system states evolution,the distributed dynamic computation offloading optimization problem in the uplinks is constructed as a dynamic stochastic game(DSG).By adopting mean field game,the DSG can be solved with two coupled partial differential equations,namely the Hamilton-Jacobi-Bellman(HJB)and FokkerPlanck-Kolmogorov(FPK)equations,and then we derive the mean field equilibrium numerically using the finite difference method.To this end,a mean field equilibrium rate offloading strategy is proposed,and based on the scenario of static channels,time-varying channels and stochastic channels,simulation results show that the cumulative cost of this offloading strategy is the lowest,and the greater the stochasticity of the channel,the greater the negative impact on the offloading strategy.