Research On Task Offloading Mechanism In Edge Computing

With the prosperous development of Mobile Internet and the rapid growth of Internet of Things,a variety of new applications are also emerging,such as mobile payment,face recognition,wearable devices,driverless,VR/AR and so on.Although the computing power of mobile terminals is getting stronger and stronger,due to power consumption,size and other limitations,they often cannot meet the computing needs of new applications.In spite of traditional cloud computing model with higher computing power,it always has a higher delay and cannot meet the needs of users.In order to reduce the delay to improve the user experience and at the same time relieve network load to a certain extent,edge computing came into being.In view of the new architecture after the introduction of edge computing,this paper focuses on the task offloading in edge computing,starting from the single-user task offloading level in dynamic environments and multi-user scenario task migration with edge server resource management.The sub-task adaptive offloading algorithm based on deep reinforcement learning(SAODQN)and the distributed task offloading algorithm based on multi-agent and load balancing(DTOMALB)are proposed,which can make reasonable offloading decisions for respective scenarios,improve user experience and balance resource utilization.The work of this paper mainly includes the following aspects:(1)In the single-user-multi-server scenario,the task of fine-grained offloading when the environment changes dynamically is studied.In view of the dynamic change of communication resources and computing resources,comprehensively considering the needs of users for delay and energy consumption,the single-user fine-grained task offloading problem is transformed into a joint optimization problem of communication and computing resources,and modeled as a markov decision process.And the algorithm named sub-task adaptive offloading based on deep reinforcement learning(SAODQN)is proposed.The algorithm can dynamically decide the offloading strategy of each sub-task during the offloading process according to the dynamic changes of the environmental state,thereby improving the user experience of the overall task.Through the simulation experiment of the algorithm,it is verified that the algorithm proposed in this paper can make a reasonable off-load decision for different users' energy consumption and delay requirements,reduce the task response time,and weigh the energy consumption of the equipment,ensuring that the scheme can bedynamically Expandability.(2)Research on task offloading and resource allocation in a multi-user single-cell scenario.In the scenario of a single cell,multiple users are connected to an edge server through a single LTE macro base station,and the edge server can schedule tasks to other edge computing servers connected thereto.Aiming at the competition and selfishness that may occur when multiple users offload their computing tasks,a global load balancing penalty factor is introduced to minimize the response time of global user tasks and make the load on each edge server relatively balanced.In addition,in view of the characteristics of dimensional explosion,scalability and poor dynamics faced by the centralized task scheduling as the number of users increases,an algorithm model for centralized training and distributed operation is proposed.By establishing each user as a Markov game model,a distributed task offloading algorithm based on multi-agent and load balancing(DTOMALB)is proposed.The simulations show that the algorithm has a certain adaptability compared to the traditional algorithm in the scenario of multi-user single cell,reduces the complexity of the algorithm compared to the centralized algorithm,reduces the average response delay of the overall user,balance the load of each edge computing server,and improve the robustness and scalability of the system.The thesis carried out simulation experiments based on the pytorch framework in chapters 3 and 4 in the second half,and compared with various algorithms.Experimental results show that the two algorithms proposed in this paper can make reasonable offload decisions in their respective scenarios.At the end of the paper,the work and achievements are summarized,and the areas where the paper can be optimized are prospected for further research.This paper includes 21 figures,7 tables and 60 references.