Research On Distributed Task Offloading And Resources Allocation For Mobile Edge Computing

Large capacity,large connection,high reliability,ultra-low latency,and green intelligence are the evolution directions of the future mobile communication systems.Mobile edge computing(MEC)technology enables the storage and computing resources to the edge of the mobile communication network,which drives the integration of data transmission,storage,and computation closer to the mobile edge devices.MEC significantly enhances the computation performances for computation-intensive and delay-sensitive applications,and it has become a hot research topic in the field of mobile communication.An important issue in mobile edge computing is how to achieve efficient computation offloading,resources allocation,and convergence of computing and networking under the dense heterogeneous wireless network and differentiated computation requirements.In response to the above problems,this dissertation studies the joint optimization methods of task offloading and multi-dimensional resources allocation in dense heterogeneous network scenarios by adopting distributed optimization theory.Under the energy harvesting enabled heterogeneous networks,the distributed methods of computation offloading,edge-cloud intelligent collaboration,and joint optimization for task offloading and resources allocation in uncertain network scenarios are proposed.The main research works are as follows:(1)Research on distributed energy harvesting enabled computation offloading technology under heterogeneous networks.In the energy harvesting enabled heterogeneous edge networks,the heterogeneity of edge network resources such as computing power,storage,and communication,as well as the time-varying network factors such as task arrival process,channel state information,and energy harvesting status,make it difficult to integrate and schedule multi-dimensional resources by adopting the traditional centralized approaches.Meanwhile,since different applications have different task characteristics and computation requirements,the traditional and rough computation offloading methods not only reduce the utilization efficiency of the scarce edge resources but also make it impossible to guarantee the requirements of different applications.To this end,the distributed on-demand computation offloading method under energy harvesting enabled heterogeneous networks is studied.Firstly,to realize on-demand edge resources allocation for the edge applications with differentiated task features and computation requirements,a dynamic bidding mechanism for edge resources based on distributed game theory is designed.Then,to ensure the stability of computation offloading performance in the long-term evolution,the multi-dimensional time coupling quantities including offloaded task,energy,and resources are decoupled by invoking perturbation Lyapunov optimization theory.Next,to choose suitable edge cloud servers rapidly,reduce communication signaling overhead,and improve computation offloading efficiency,a computation offloading pre-screening criterion is designed by balancing the task queue backlog,energy harvesting level,and task offloading cost.Finally,a distributed energy harvesting enabled computation offloading algorithm in heterogeneous edge networks is proposed.Theoretical analysis and simulation results show that the proposed scheme can effectively improve energy efficiency and reduce computation latency while achieving Stackelberg equilibrium.(2)Research on the edge cloud intelligent collaboration and task offloading optimization technology in multi-agent network scenarios.The autonomous and random edge cloud access of users will cause insufficient and unbalanced utilization of edge resources,as well as reduce edge cloud resources utilization and user computation experience.To this end,the edge cloud intelligent collaboration and task offloading optimization method are studied in heterogeneous networks.Firstly,the heterogeneous edge cloud and energy harvesting enabled mobile devices are regarded as agent units that can independently make decisions.Considering the actual computation offloading scenarios,the edge cloud resources allocation and the task offloading of mobile devices is abstracted into a typical distributed Stackelberg game model,in which the edge cloud acts as the leader to allocate computing and transmission resources,and the mobile devices act as the followers to offload task.Then,considering the stochastic time-varying edge environment and incomplete state observation,the problem of developing strategy for the agents is modeled as a partially observable Markov decision process,and a multi-agent deep deterministic policy gradient model based on Stackelberg dynamic game is designed to learn the optimal edge cloud resources allocation and task offloading strategies.Finally,the simulations prove that the proposed algorithm can effectively improve the computation offloading successful rate of edge cloud,and reduce the task drop rate of mobile devices.(3)Research on the joint optimization method of task offloading and resources allocation in uncertain edge cloud network scenarios.In the dense and heterogeneous wireless edge networks,the randomness and time variability of task arrival,wireless channel states,user movement,and energy harvesting result in an uncertain edge network environment,which poses serious challenges to design efficient task offloading and resources allocation strategies.To this end,the joint optimization method of distributed task offloading and resources allocation in uncertain heterogeneous network scenarios is studied.Firstly,considering the impact of stochastic task queuing delay in cloud servers and user mobility on computation offloading,a dynamic task offloading and edge cloud resources adaptive allocation model based on computing and networking coordination is designed.Secondly,to improve computation offloading performance in multi-dimensional uncertain factors such as user mobility and task queuing delay in the time-varying network environment,a joint optimization method of task offloading and resources allocation based on distributed multi-stage stochastic programming theory is proposed,which can take a posteriori resources allocation to compensate for prediction inaccuracies caused by uncertain network environments.Finally,the simulations prove that the proposed method can reduce the cost of task offloading,and effectively improve the success rate of computing offloading.