Research And Improvement On Dynamic And Static Resource Scheduling Strategies Based On Kubernetes

With the continuous development and widespread application of container technology,Docker containers have become the preferred choice for enterprises to deploy cloud services due to their lightweight,flexible,fast startup,and low resource consumption advantages.In order to efficiently manage and schedule numerous containers in cloud platforms,container orchestration systems have emerged.Kubernetes,as an open-source container orchestration system based on Docker,with its powerful container management capabilities,provides users with stable and convenient services,becoming an industry standard for deploying containerized applications.However,Kubernetes’ resource scheduling strategy still has shortcomings.To further improve the resource utilization and service quality of Kubernetes clusters,this article aims to address the shortcomings of the default resource scheduling strategy.The main work is as follows:(1)A static resource scheduling strategy based on improved artificial bee colony algorithm is proposed to address the issue of Kubernetes’ static scheduling strategy considering limited resource indicators and inability to consider cluster load balancing from a global perspective.First,build a multi pod resource scheduling model,add CPU,memory,bandwidth,disk IO resource indicators to the indicator evaluation system,and design a fitness function with task time consumption,cluster resource utilization and load balance as variables.Secondly,the artificial bee colony algorithm is improved by using adaptive neighborhood search,chaotic factors,dynamic weights,and beneficial information of the optimal solution to improve convergence accuracy and speed.Finally,an improved artificial bee colony algorithm is used to drive a multi pod resource scheduling model,which performs pod scheduling from a global optimal perspective,improving the load balance and resource utilization of the cluster.(2)A predictive elastic scaling strategy based on predictive models is proposed to address the issue of delayed response when Kubernetes’ default responsive elastic scaling strategy faces load changes.Firstly,design the overall architecture of Kubernetes,including monitoring module,prediction module,and scaling module.Secondly,a combination prediction model of ARIMA prediction model and SVM regression model is constructed,which uses ARIMA prediction model to predict linear data in time series data,and then uses prediction residual as input to SVM regression model.The prediction results of the two models are combined to reduce prediction errors,and the prediction load and current load are comprehensively considered to achieve the expansion and contraction of Pod replica.Finally,by setting Pod priority,the scaling strategy is improved to improve cluster load balancing.(3)In response to the problem of considering fewer indicators for Kubernetes elastic scaling strategy threshold,a scaling strategy with comprehensive load as the threshold is proposed.Incorporate CPU,memory,network bandwidth,and disk IO into the comprehensive load,and use a combination of Analytic Hierarchy Process and Grey Correlation Method to assign weights to various resource indicators,taking into account subjective experience and objective reality,to make the setting of resource indicator weights more reasonable.The experiment shows that the improved static resource scheduling strategy effectively improves the resource utilization and load balance of the cluster;The elastic scaling strategy based on hybrid prediction model can effectively shorten request response time and improve the service quality of applications,and the improved scaling strategy effectively improves the load balance of the cluster after scaling.