Research On Task Scheduling And Container Deployment Strategies For Heterogeneous Edge Cluster

Under the context of Industry 4.0,with the continuous development and popularization of technologies such as the Internet of Things,mobile Internet,and artificial intelligence,there is a growing demand for low latency,high bandwidth,and high reliability from users.Edge computing,as a new computing paradigm,has emerged.However,with the development of edge computing and the increasing number of edge devices,the heterogeneity of edge nodes has become an inevitable challenge in edge computing systems.The heterogeneity of edge environments poses two urgent problems for task scheduling and service deployment: 1)how to schedule tasks and allocate resources in a heterogeneous edge environment to ensure efficient task execution;and 2)how to deploy diverse services reasonably in the face of cluster heterogeneity,so as to improve deployment performance and reduce task processing time.This thesis focuses on two aspects of research: task scheduling algorithms and service deployment strategies in heterogeneous edge environments.It proposes a performance optimization scheme for heterogeneous edge environments,which includes the following two parts:(1)In order to improve task execution performance,this thesis proposes a parallel task scheduling algorithm based on non-cooperative game theory.Starting from the resource requirements of tasks,resources are divided into soft resources and hard resources,and the impact of soft resources on task execution performance is explored,and a task performance estimation model is established.Then,based on this model,the impact factor of tasks on nodes is calculated,and resources are allocated for parallel tasks based on this.Finally,combining the task performance estimation model and resource allocation algorithm,task scheduling is viewed as a non-cooperative game problem,and a parallel task scheduling algorithm based on non-cooperative game theory(PTSA-NE)is proposed.In comparative experiments with random scheduling,round-robin scheduling,and weighted round-robin scheduling algorithms,PTSA-NE improved average task performance by 24.1%,119.8%,and 90.10%,respectively,and average system performance by 69.13%,86.54%,and 240.40%,respectively,and reduced average task latency by 42.38%,53.60%,and 62.60%,respectively.In addition,the extended experiments show that PTSA-NE has good scalability.(2)In order to improve service deployment performance,this thesis proposes a container deployment algorithm based on relative resource requirements and load updates(RRLA),which consists of two parts.Firstly,the heterogeneity of nodes and the relative resource requirements of containers are characterized,and a container predeployment algorithm based on relative resource requirements(RRDA)is proposed,which uses a greedy approach to maximize the pre-deployment performance of containers.Then,in order to quantify the workload of containers during runtime,container load and load update methods are defined,and a container deployment adjustment algorithm based on load updates(LLUA)is proposed.LLUA can track the load of containers and timely migrate high-load containers to higher-performance nodes to improve deployment performance.In comparative simulation experiments with Kubernetes,Swarm,and ECSched,RRLA effectively reduced task completion time and task latency.When the number of tasks was 1000,RRLA reduced the task sequence completion time by 60.95%,64.43%,and 70.62%,respectively,and the average task latency by 45.99%,43.76%,and 57.15%,respectively.Finally,the effectiveness of the LLUA algorithm was verified by adding the LLUA mechanism to Kubernetes,Swarm,and ECSched.