| With the rapid rise of cloud native technology,the trend of enterprises embracing containerization will continue.When faced with the challenges of managing containers,a powerful open source container management system called Kubernetes is rapidly gaining the favor of many enterprises due to its mobility,scalability,and self-healing.In order to take full advantage of enterprise facilities resources,ensure the rational allocation of cluster resources,and solve the problems of difficult application deployment and low work efficiency of employees,this paper uses the existing infrastructure of the enterprise,designs and implements a container cloud platform with independent optimization of resource allocation based on Kubernetes,which provides a safe,reliable and convenient cloud service platform for the enterprise.In the resource-sharing container cloud platform,how to use resources efficiently and reasonably is a problem worthy of attention.Firstly,this paper analyzes the current situation of unfair resource allocation of multi task queues in Kubernetes.Aiming at the shortcomings of Kubernetes scheduler default sorting algorithm,dominant resource fairness algorithm DRF and its related extension algorithms,a weighted multi dominant resource fairness algorithm WMDRF is proposed.The algorithm considers multi-dimensional resources and has the properties of Pareto optimality,policy proof and no jealousy.After experimental analysis,WMDRF can start more tasks than DRF under the same resource constraints,and the greater the difference in task resource requirements between users,the better the performance.Secondly,aiming at the problem of low utilization of CPU resources in cluster applications,this paper analyzes the shortcomings of Kubernetes automatic scaling,and designs a hybrid prediction model of ARIMA and Holt based on STL time series decomposition combined with the advantages and disadvantages of different prediction methods.This model is used to predict the CPU resource usage of applications in advance.After experimental analysis,the hybrid model has good stability,prediction accuracy and prediction speed,which can control 91.01%of the prediction result error within 5%.Compared with ARIMA,Holt-winters and ARIMA+triple exponential smoothing hybrid method,the average prediction error is reduced by 36.08%,41.84%and 32.05%respectively,and the prediction speed is also 5.87%faster than ARIMA.After the requirement analysis and outline design,this paper designs and implements each functional module of the container cloud platform with independent optimization of resource allocation in detail.Including basic functions such as user registration,node control,application installation,and independent optimization of resource allocation.In the user management module,aiming at the problem of unbalanced allocation of user resource quota,a resource dynamic allocation mechanism is proposed.The resource quota is dynamically controlled according to the user resource use intensity.Based on this mechanism,the function of independent optimization of user resource quota is designed and realized;In the cluster management module,the independent optimization function of Queue task resource allocation in the cluster is designed and realized based on WMDRF;In the application management module,the independent optimization function of application resource limitation is designed and realized based on the hybrid prediction algorithm.Finally,this paper tests and analyzes the functions of each module of the container cloud platform in a real cluster environment.The operation effects of various functions meet the expectations,and the independent optimization function of resource allocation in different modules also achieves the design purpose:the optimization function of user resource quota can make the resource quota allocation of platform users more reasonable and balanced;The optimization function of task resource allocation enables the fair and efficient allocation of limited resources among different user tasks;The optimization function of application resource limitation increases the application CPU utilization by 18.08%on average. |
