Research And Implementation Of Resource Scheduling Strategy In Container Cloud

With the gradual maturity of the Cloud Native concept and the rapid development of its related technologies,the deployment of applications on the cloud has become fast and efficient.Once containerized,the application can be quickly deployed to the cloud platform,and container management can then be performed by using container choreography tools.Therefore,Deploying applications to cloud platforms has led to widespread use of container choreography system Kubernetes.However,Kubernetes scheduling mechanism and elastic scaling strategy have the following shortcomings.First,I/O pod resource objects are not considered in pod resource scheduling.Second,pod resource scheduling does not consider the actual load of the cluster.Third,the load balancing scheduling algorithm only considers CPU and memory computing resources without other types of resources and node performance.Fourth,elastic scaling strategy HPA performs elastic scaling based on a single judgment and cannot detect the change of cluster load.Aiming at the above four problems,this thesis takes v1.21.5 version Kubernetes as the research object and makes the following work:Firstly,in view of the first and second points above,this thesis implements a custom monitoring component based on the open source tool Prometheus to monitor the real-time load of the cluster and obtain real-time load data.A cluster I/O balancing based scheduling algorithm,Balanced IOResource Priority,is proposed.It combines real-time data obtained by monitoring components related to cluster disk I/O and network I/O with disk requests for Pod resources to be scheduled,and is used for Pod resource object scheduling.According to the third problem,a dynamic weight balanced scheduling algorithm,Balance LoadPerformance Priority,is designed based on the real-time load of nodes and the performance of nodes.Which node load indicators include CPU,memory,disk I/O,and network I/O,node performance is measured by static indicators including CPU frequency,number of CPU cores,memory capacity,disk I/O upper limit and network I/O upper limit.The weight of each index is determined by Analytic Hierarchy Process.Aiming at the fourth problem,a new elastic scaling method based on Statistical Process Monitoring And Control and Western Electric Rules is proposed.The SPC is used to automatically calculate the average value,upper and lower threshold value of monitored resource objects in the cluster,and the threshold range is determined by ”Six Sigma” method.WER is used to enrich the judgment basis of cluster elastic scaling signal events.Finally,a Kubernetes cluster is built to test the proposed Balanced IOResource Priority algorithm,Balance Load Performance Priority algorithm,and WER-based elastic scaling strategy.The experimental results show that the Balanced IOResource Priority algorithm can balance the use of cluster I/O resources better,the Balance Load Performance Priority algorithm can improve the overall balance efficiency of the cluster,and elastic scaling of the custom elastic scaling component is more sensitive.