Prediction And Configuration Optimization Research For Kubernetes Container Cloud Resources

Cloud computing as a high-efficiency,low-cost computing model has become a classic distributed architecture.The container technology represented by Docker and Kubernetes has rapidly replaced the traditional virtual machine solution in cloud computing as the first choice of users by virtue of its flexibility and efficiency.As more and more applications are deployed in containers,the problem of resource management optimization under large-scale container cloud clusters also comes to the fore.The finegrained,intelligent development and operation concepts such as Dev Ops and AIOps are also gaining popularity,and the requirements for high elasticity and flexibility of resources are getting stronger and stronger.How to efficiently and reasonably utilize resources under container cloud clusters has become an important part of current research in the field.Accurate prediction of container cloud resources is the basis for flexible scheduling of resources,which can make the container cloud platform respond to the changes of resources in the cluster in advance and realize the accurate scheduling and allocation of resources.The research on the configuration optimization of container cloud resources can better ensure the efficiency and rationality of resource utilization.Therefore,this paper conducts relevant research for Kubernetes container clusters from two aspects: cloud resource prediction and cloud resource allocation optimization,as follows.1.This paper proposes a hybrid neural network-based multivariate cloud resource prediction model,which uses multivariate cloud resource data for information extraction,model construction and fusion prediction.Firstly,constructing multiple channels using different features in multivariate cloud resource data.Secondly,fusing the spatio-temporal feature extraction ability of CNN,the sequence information capturing ability of Bi LSTM,the key factor screening ability of Self-attention for prediction.Finally,according to the dynamic characteristics of cloud resources and actual prediction requirements,weights are designed for fusion prediction using Analytic Hierarchy Process.The prediction results of the model are more consistent with the original series and the prediction error is smaller2.This paper proposes a Stackelberg model based cloud resource allocation optimization algorithm for cluster scheduling from a macroscopic perspective.Firstly,obtain dynamic cloud resource change information and service status information through inspection tools.Secondly,a number of resource metrics are customized as load balancing metrics,and the evaluation method of service quality is given.Finally,a Stackelberg model with service quality as the leader and load balancing as the follower is established to obtain the selection strategy of the optimal nodes to reduce the load balancing of the cluster and utilize the fragmented cloud cluster under the premise of guaranteeing service performance resources.Deploying the algorithm in a Kubernetes cluster,the algorithm dramatically improves the average quality of service of the cluster and the average load balance of the cluster,increases the average utilization of each resource,and reduces the fragmentation of the cluster resources.