Research On Edge Task Unloading And Resource Allocation Method Based On Energy Consumption Perception

Compared with the traditional way of data processing in the remote operation center,edge computing(EC)provides a network environment with high bandwidth and low delay by sinking the computing and storage capacity to the network edge close to the terminal equipment,to improve the service quality of delay sensitive services.As one of the key technologies of edge computing,edge computing offload technology offloads the tasks running on the terminal equipment to the edge server through reasonable offload decision and resource allocation strategy,uses the server’s computing and storage resources to complete the task execution,and reduces the task completion delay and energy consumption of the terminal equipment.In addition,due to the wide range of deployment locations and large number of edge servers,the energy consumption generated at the edge server is usually very considerable when performing edge tasks.Therefore,it is particularly important to design appropriate task unloading strategies for different task unloading scenarios to optimize the energy consumption of the edge system.Compared with the traditional way of data processing in the remote operation center,edge computing(EC)provides a network environment with high bandwidth and low delay by sinking the computing and storage capacity to the network edge close to the terminal equipment,to improve the service quality of delay sensitive services.As one of the key technologies of edge computing,edge computing offload technology offloads the tasks running on the terminal equipment to the edge server through reasonable offload decision and resource allocation strategy,uses the server’s computing and storage resources to complete the task execution,and reduces the task completion delay and energy consumption of the terminal equipment.In addition,due to the wide range of deployment locations and large number of edge servers,the energy consumption generated at the edge server is usually very considerable when performing edge tasks.Therefore,it is particularly important to design appropriate task unloading strategies for different task unloading scenarios to optimize the energy consumption of the edge system.However,the energy consumption oriented research in the edge computing scenario still has the following problems:1)for the energy consumption optimization of the edge server,the existing research lacks a mathematical model that can accurately evaluate the energy consumption of the edge server,and the existing model does not consider the startup energy consumption of the edge server and the energy consumption of the switch on the transmission link.2)For the end side hybrid edge unloading scenario,the existing research mostly focuses on the research of task execution strategy.When the task is executed locally,there is a lack of reasonable computing resource scheduling strategy to further optimize the energy consumption of terminal equipment and prolong the service life of terminal equipment.To solve the above problems,aiming at the edge task unloading scenario involved in question 1),this paper first constructs an energy consumption model that can accurately evaluate the energy consumption of the edge network.After considering the computational resource constraints and bandwidth constraints,the task unloading problem is expressed as an optimization problem model,Then,based on PPO(near end policy optimization),an edge task unloading algorithm for energy consumption optimization is proposed.Aiming at the edge unloading scenario of terminal edge combination involved in question 2),this paper first constructs the overhead model of local execution and edge and proposes an edge task unloading algorithm based on DQN(Deep Qlearning).In addition,when the task is executed locally,this paper also proposes a terminal CPU frequency scheduling strategy to further optimize the terminal energy consumption.Finally,the proposed algorithm is simulated to verify the effectiveness of the above algorithm and model.