Research On Terminal Hierarchical Resource Management Method Oriented To The Application Combining Human Machine And Thing

Nowadays,the rapid development of Internet of Thing has made the integration and symbiosis of human,machine and thing an irreversible trend.Currently,the development of integration of human,machine and thing still faces a lot of challenges.How to effectively carry out resource scheduling under edge scenarios is one of the key problems.Huge amounts of data are produced by edge devices.Considering the heavy burden of network bandwidth and the service delay requirements of the delay-sensitive applications,processing the data at the network edge is a great choice.However,edge devices usually have a lot of limitations on computational capacity and energy which will severely influence the quality of service.An effective and efficient offloading strategy is the key point to address this issue.But when devices have different capacities and tasks become complex in terms of density and size,traditional offloading strategies based on Game Theory or Operation Research have lost the effectiveness when facing complex tasks and heterogeneous edge devices.A self-adaptive original strategy is much needed for device management and computational resource scheduling.Focusing on this challenge,regional resource scheduling is of great importance.Thus,in this paper a double offloading framework is proposed to simulates the offloading process in real edge scenario which consists of several edge servers and devices.The local processing layer is responsible for generating tasks,uploading and processing a part of tasks.The offloading layer receives these tasks,sorts and processes them which effectively reduces the computation load of the local processing layer.At the same time,double offloading framework guarantees the resource utilization rate and task processing efficiency of offloading layer by utilizes idle resources in other regions.The offloading process is formulated as a Markov Decision Process(MDP)and a deep reinforcement learning(DRL)algorithm named asynchronous advantage actor-critic(A3C)is utilized as the offloading decision making strategy to balance the workload of edge servers and finally reduce the overhead in terms of energy and time.A comprehensive task data set with seven different task densities is created to simulated the real task production in edge computing.Comprehensive comparison experiments between local computing,wide-used DRL algorithm Deep Q-Network,A3 C,and double offloading framework are conducted using the task benchmark in a long sequential time period.The results show that double offloading framework with A3 C algorithm performs much better on self-adjusting and overhead reduction than other methods.The quality of services of edge application have been obviously improved.