Research On Resources Scheduling Strategy Of Container Cloud Platform Based On Kubernetes

Container technology,like Docker is easy and practical to develop,publish and deply the application,so it has attracted broad attention in the industry.Due to its focus only on providing container and image,an integrated container cloud management platform is required to efficiently complete a series of tasks such as container scheduling deployment,resource scheduling,service discovery and health monitoring.Kubernetes is the preferred choice in the container cluster scheduling system,but its resource scheduling strategy and scheduling algorithm are too simple to deal with complex applications.The resource scheduling modle is improved and redesigned after the deep study of Kubernetes technology in this thesis.The main contents are as follows:1.Kubernetes resource model is improved.Kubernetes only consider the influence of CPU and memory in resource scheduling,but if Pod works well,the data between image download and persistent storage volume are required to exchange.Compared the original one,the new model in this thesis views the image download speed and data transmission speed as the consideration factor in resource scheduling as well.2.Kubernetes user binding strategy is improved.Its user binding strategy is comparatively simple and strong,therefore a weak binding strategy is formulated based on it in this thesis and meanwhile the matching rules for user binding is enhanced,which supports the matching period detection for expressions.3.Preemptive scheduling strategy is designed.Pod is divided into three priorities according to the restart policy.When the host resource is insufficient,Pod with high priority can preempt the low priority resources in order to greatly raise the running ratio of Pod with high priority.4.Dynamic load balancing scheduling stragey is designed.Kubernetes scheduling strategy is static,which means that Pods are not moved from scheduled to node to its end of lifecycle.A dynamic load balancing algorithm based on Kubernetes is formulated by combining the improved resource model and preemptive user binding strategy.Static scheduling of this algorithm is responsible for dispatching each Pod of the Pod queue to the node of most suitable resource description file;and dynamic scheduling regularly provides the running situation of host computer and Pod through the monitor to the scheduler,and then the scheduler makes dynamic scheduling in accordance with the overall load of system,so that Pod on the higher load node is moved to the low load node to maintain the overall load balancing.