Reserarch On Resource Scheduling Based On Kubernetes Cluster

In today’s rapidly evolving cloud computing environment,container orchestration tools play an indispensable role as one of the important supporting technologies for containerization technology.By significantly simplifying the resource management and scheduling distribution process in cloud platforms,container scheduling tools achieve efficient resource utilization and provide powerful support for cloud computing.The scheduling algorithm has become a key research topic in the field of container orchestration for the key issue of real-time response to resource requests in container clouds.Therefore,it is of great importance to design efficient resource scheduling algorithms.However,the default scheduling algorithm used by the current mainstream container orchestration tool Kubernetes has two main problems that need to be further studied and improved to meet the demand for efficient resource scheduling in the cloud computing environment.First,the scheduling algorithm of the scheduler only refers to two metrics,CPU and memory,during the compute node scoring process,which cannot meet the needs of other types of Pod applications;second,the overall load balancing problem of the cluster after deploying a large number of Pod applications is also not sufficiently considered.To solve the above problems,the main research of the paper has two aspects.The scheduling algorithm provided by Kubernetes only refers to CPU and memory resources as scoring indicators for computing nodes in the optimization stage,while ignoring that different types of Pods have different proportions of node resources,which can easily lead to a performance bottleneck for a certain resource,resulting in an imbalance in resource utilization by nodes.Therefore,this paper proposes an improved bald eagle search algorithm(TBESK)to optimize the existing Kubernetes scheduling algorithm.First,the Kubernetes performance indicator is extended on the basis of the original resource indicator,and then the pods waiting to be created are divided into compressible consumption type,incompressible consumption type,and balanced type,and corresponding weights are set for each type of Pod.Then optimize the bald eagle algorithm through t distribution and Tent mapping,and apply it to the improved TBESK method to select the appropriate deployment node.This method can meet the needs of multiple types of Pod applications,while effectively balancing cluster node resources under heavy cluster load.(1)The scheduling algorithm comes with Kubernetes only refers to two resources,CPU and memory,as the scoring index of computing nodes in the preference stage,and also ignores the fact that different types of Pods occupy different proportions of node resources,which easily leads to a certain resource reaching a performance bottleneck,thus causing the problem of imbalance of resource usage by nodes.In this paper,we propose an optimization of the existing Kubernetes scheduling algorithm based on an improved bald eagle search algorithm(TBESK).First,the Kubernetes performance metrics are extended on the basis of the original resource metrics,after which the Pods waiting to be created are distinguished into compressible consumption type,incompressible consumption type,and balanced type,and the corresponding weights are set for each type of Pods.Then the Bald Eagle algorithm is optimized by t-distribution and Tent mapping and applied to the improved TBESK method to select the appropriate deployment nodes.The method is able to meet the demand of multiple types of Pod applications,and at the same time,it can effectively balance the cluster node resources in the case of high cluster load.(2)For Kubernetes clusters that do not consider the changes of real-time resources of nodes when deploying Pods may affect the normal use of applications in the Pods,this paper proposes a resource balancing algorithm based on real-time resources of applications(SFPK)by combining the Scheduling Framework scheduling framework and Prometheus.The algorithm first collects real-time resource metrics of nodes through Prometheus.Then a custom scheduling logic is built to implement scheduling of Pods based on the real-time resource metrics of nodes to improve the node selection strategy of the default scheduling algorithm.The experimental results show that the TBESK algorithm can meet the demand of multi-class Pod applications and effectively balance the cluster node resources under the large cluster load compared with the default scheduling algorithm of Kubernetes;SFPK can better prevent Pods from being killed due to node resource overload,thus better ensuring the overall high availability of the cluster.