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

With the development of container virtualization technology,docker container has rapidly become the first choice for enterprises to deploy cloud platform with the advantages of lightweight,fast startup,security isolation and low resource consumption.There are many containers in the cloud platform,so an efficient container choreography system is needed to manage the cluster.Among the numerous products in the market,kubernetes,a representative open source tool,is the leader of the new generation.Its excellent performance can provide more convenient services for the majority of users.This thesis studies the theory of kubernetes cloud platform,introduces its system architecture and function design,and analyzes the scheduling mechanism of kubernetes system integration.In the process of cloud platform operation,the resource demand will be in a constantly changing process when the cloud load fluctuates.The original fixed resource allocation method is lack of flexibility,and the waste of resources is serious.To solve the above problems,this thesis designs a set of resource scheduling optimization strategy.Firstly,a resource scheduling mechanism based on prediction model is proposed to solve the lag problem of current static resource scheduling schemes.After observing the resource usage characteristics of cloud platform,it is found that the resource usage will change with the fluctuation of application load,including linear and nonlinear characteristics.Therefore,the ARIMA model suitable for linear sequence forecasting and the BP neural network suitable for predicting nonlinear sequences are studied,and a combined forecasting model is proposed by combining the two forecasting models.Secondly,kubernetes scheduler cant schedule the resource before the bottleneck of resource consumption,which is easy to cause the bottleneck of resource consumption.In order to ensure the quality of service of resource scheduling in kubernetes system,the results of composite prediction model are applied to the resource scheduling module,so that the kubernetes cluster application can dynamically select the scheduling scheme.Finally,the existing kubernetes infrastructure is optimized by adding four modules:monitoring module,prediction module,auto scaling module and resource scheduling module.In order to verify the effect of the improved Kubernetes cloud platform,this thesis conducted a series of experiments on the accuracy of the prediction model,the automatic scaling strategy and the resource scheduling strategy.The experimental results show that the proposed combined forecasting model has relatively small errors and can better reflect the changing law of application resource requests;The automatic scaling strategy can expand the cluster size with the increase of load,and improve the availability of the cluster;Resource scheduling mechanism can more effectively improve the resource balance of the application and improve the service quality of the application.