Research And Implementation Of Heterogeneous Resource Scheduling Based On Kubernetes

In recent years,with the rapid development of big data,artificial intelligence and other fields,the demand of related applications for computing resources and storage resources is growing.In order to support these growing demands,more and more applications are migrating to the"cloud".Cloud computing is based on the Internet,unified management and scheduling of computing resources,storage resources,software resources,etc.,and these resources are shared to end users according to the demand.In addition,cloud computing platform can reduce the development difficulty of developers,improve the development efficiency,meet the needs of the whole stack,and achieve fast and stable iteration.The emerging virtualization technology docker container is used by more and more enterprises to deploy cloud services to actual production because of its fast startup and low resource consumption.Kubernetes is an open source container management system based on docker container.With its advantages of open source,scalability and low deployment difficulty,Kubernetes has become the preferred solution for container choreography.However,in the actual production,Kubernetes still has shortcomings in resource scheduling,which can not meet the needs of users in heterogeneous and complex environment.Therefore,this paper proposes a heterogeneous resource scheduling scheme based on Kubernetes.The specific work is as follows:This paper designs the scalable scheduling strategy and balanced scheduling strategy for CPU,GPU,memory and other heterogeneous resources.(1)This paper proposes a resource scheduling method based on historical load data,which can improve the problem of low resource utilization in low load state caused by experience allocation.(2)This paper proposes a GPU elastic scaling solution based on priority algorithm,which can improve the utilization of GPU resources in the case of cluster GPU resource shortage.(3)This paper proposes a resource balanced allocation model,which can avoid the high utilization rate of one resource and low utilization rate of other resources in the cluster with a variety of heterogeneous resources,and achieve the effect of fair allocation of cluster resources,so as to improve the utilization rate of resources.Experiments show that the above three schemes can improve the resource utilization.