Scheduling Optimization For Kubernetes Container Management

With the extensive application of micro-service architecture and container technology in cloud computing,managing containerized micro-services in large clusters has become a goal that developers are eager to achieve.Although many companies,such as Amazon,Red Hat and VMware,have launched its own PaaS layer cluster management system,the development of PaaS still evolves rather slowly due to some intrinsic limitations.Recently,the emergence of Kubernetes has broken the ice,and showed a thriving development.It has become the preferred container cluster management platform for AWS,Oracle,and Pivotal.However,Kubernetes can only perform simple scheduling management operation,which prevents Kubernetes from meeting the various and complex requirements of special application scenarios.Based on the thorough analysis of Kubernetes scheduling framework and default scheduling scheme,this thesis designs a set of reasonable and effective scheduling scheme,optimizing Kubernetes scheduling mechanism from three stages,respectively pretreatment,decision-making,runtime stage.(1)Pretreatment scheduling mechanism: adding node ready state controller module on the basis of Kubernetes original module solves the problem that the original scheduling scheme cannot detect the state of daemon process or the scheduling state of managed node.Based on the Limits/Requests scheduling rules,Pod resources guarantee and node resource oversold scheduling scheme was designed,not only avoiding the downtime caused by the shortage of resources,but also making the most of node to calculate resources.(2)Decision-making scheduling mechanism: adding two nodes presele ction algorithm and two nodes priorities algorithm on the basis of th e original Kubernetes scheduling algorithm improves the scheduling pe rformance under the application scenarios of affinity and mutual excl usivity scheduling rules.(3)The adaptive adjustment mechanism during running: adding passive scheduling protection controller module on the basis of original Kubernetes scheduling algorithm ensures the stable operation of the core business Pod when the node proceed management operation.The passive scheduling scheme also designed the signal quantity and threshold value triggered by the secondary scheduling,guaranteeing the stability of the nodes in the cluster as much as possible.The above scheduling scheme was verified and made contrastive analysis through experiment.The results show that the improved scheduling scheme can improve the scheduling efficiency,enhance the usability,and better meet the scheduling demand in the practicalapplication scene.