Research On QoS Optimization Technologies Of Dynamic Resource Scheduling In Edge Computing

With the rapid development of the Internet of Things(Io T),the widespread popularity of wireless networks,and the increasing demand for communication and computation from smart terminals and mobile applications,the edge computing paradigm has emerged.By moving the services and functions in the cloud computing center near users,edge computing can provide powerful communications,storage,and computing capabilities.Dynamic resource scheduling in edge computing is one of the key issues of edge systems,which has attracted increasing attention from scholars.In edge computing,mobile users can offload their delay-sensitive or computation-intensive applications to edge servers for processing to obtain higher Quality of Service(QoS).On the one hand,the offloading process involves the competition and allocation of various computing,communication,and storage resources in the edge system.On the other hand,edge resource providers need to consider how to provide edge resources to satisfy users' requirements.However,due to the diversity of users' requests in the edge system,and the limited,decentralized,and heterogeneous nature of edge resources,the resource scheduling problem in edge computing becomes more and more difficult.To this end,this thesis mainly focuses on how to schedule edge resources dynamically to optimize QoS.Furthermore,taking the Io T and the Internet of Vehicles(Io V)as typical edge computing scenarios,this thesis study the problems of joint request offloading and resource allocation,workload prediction and resource allocation,and collaborative task scheduling.The original contributions of this thesis are as follows.(1)Dynamic request scheduling optimization under the coordination of multiple base stations.The phenomenon of multiple access interference under the 5G communication protocol,as well as challenges such as joint request offloading and resource allocation in multi-user-multi-edge make it difficult to design a power allocation scheme and dynamic request scheduling algorithm.To address this problem,this thesis first proposes a subgradient-based non-cooperative game model for power allocation to minimize transmission energy consumption by using pseudo-convex technology.Second,the joint request offloading and resource scheduling problem is modeled as a mixed-integer nonlinear program to minimize the response delay of requests.The problem is analyzed as a double decision problem.Finally,this thesis presents a multi-objective optimization algorithm based on the elite non-dominated sorting genetic algorithm.The simulation results show that the proposed power allocation algorithm can effectively save transmission energy compared with other similar algorithms.In the dynamic edge system,the proposed request offloading and resource scheduling algorithm outperforms the existing methods in terms of response rate and can maintain a good performance.(2)Dynamic resource provisioning optimization algorithm based on multiple service work-load prediction.The dynamic diversity of user requests in the Smart and Connected Communities(SCC)and the limited edge resources make the research on workload prediction and resource provisioning scheme full of challenges.First,the advantages and challenges of containerized edge computing are demonstrated through system experiments.A containerized edge computing framework for resource provisioning is proposed,integrating workload prediction and resource pre-provisioning.Second,based on Fast Fourier Transform(FFT),an online periodic workload prediction algorithm is designed.Finally,a resource provisioning algorithm based on a first-order auto-regressive model and a proportional-integral controller is proposed according to the predicted workload distribution.The algorithm is a self-adaptive controller to tune the resource for containers.Simulation experiments show that the proposed workload prediction algorithm has higher prediction accuracy than the other two prediction algorithms.The experimental results of the testbed show that,compared with the baseline algorithm,the control-based resource pre-provisioning algorithm has lower service delay and higher resource utilization.(3)Distributed task scheduling optimization algorithm based on multi-agent reinforcement learning.To address the problems of overloaded computing tasks and missing services of single Road Side Unit(RSU)in Vehicular edge computing(VEC),this thesis studies distributed collaborative task scheduling based on multi-agent deep reinforcement learning.First,the interactions involved in task scheduling among distributed RSUs is modeled as a Markov game.Second,given that multi-agent deep reinforcement learning is a promising approach for the Markov game in decision optimization,this thesis proposes a collaborative task scheduling algorithm based on multi-agent deep reinforcement learning for VEC,aiming to minimize the long-term average delay of tasks.Based on the distributed deployment and collective training architecture of counterfactual multi-agent policy gradient,the proposed algorithm adopts the advantage of the action semantics network to facilitate cooperation among multiple RSUs.The evaluation results of both testbed and simulation demonstrate the effectiveness of our proposed algorithm.Compared with baselines,the proposed algorithm can converge faster and achieve lower average task delay.Moreover,simulation results in different scenarios show that the proposed algorithm has good scalability and robust stability.(4)Greedy-based scheduling optimization algorithm based on task dependence in mobile vehicle clusters.The phenomenon of under-utilization of vehicle resources in the Io V system and the task dependence in delay-sensitive applications have motivated this thesis to study vehicle aggregation and scheduling of task-dependent applications.To verify the feasibility of aggregating vehicular resources in the real world,this thesis first gives several observations based on the analysis results of the real traffic dataset.Then,a latencyaware real-time scheduling framework for the edge-enabled Io V is designed.Mobile users can offload applications to the scheduling framework,and the offloading tasks can be scheduled to the appropriate vehicular resources in real-time.First,this thesis proposes a clustering-based algorithm to generate vehicle clusters,which treats connected vehicles as edge computation resources to provide cooperative computing services.Second,considering the dependency relationship between tasks in the job,this thesis presents a greedy-based task scheduling algorithm for offloading jobs,the objective of which is to minimize the total latency of the job as well as maximize the resource utilization of vehicle clusters.The simulation experiment based on the real traffic dataset shows that vehicle clusters generated by the proposed clustering-based algorithm can maintain a stable period to provide computing service.And,the experiments on the testbed include two case studies demonstrate that the proposed scheme is better than baselines in terms of delay and resource utilization.