Design And Implementation Of Container Auto Scaling Algorithm Platform Based On Kubernetes

In the development of micro-service,developers find that the availability of the system decreases because of the dependence on the configuration management of the server itself,which makes the efficiency of developing micro-service mode for a faster software iteration not rise but fall.The rise of container technology has solved problems such as configuration management.Firstly,the critical concept and system architecture of container technology and container scheduling management platform Kubernetes are demonstrated.On this basis,the realization principle of Kubernetes' automatic scaling technology is introduced.Also,by combining with business scenarios,the shortcomings of current automatic scaling strategy in actual production scenarios are discussed.The optimization solution of scheduling strategy is proposed to solve the performance bottleneck problem caused by the inability to meet the demand quickly.In the selection of load prediction model algorithm,by analyzing and comparing their advantages and disadvantages of the suitable scenarios of mobile average algorithm,linear regression prediction algorithm and differential autoregressive mobile average prediction algorithm,combined with actual business scenarios,the load request situation in the future is predicted so that the business system can expand ahead of the peak load of the request so that reducing the request response time to improve user experience.On the other hand,the cloud services provided by public cloud adopt the mode of charging by time.This paper focus on solving the problem that the dynamic scaling strategy of Kubernetes can't optimize the policy control of the underlying resource pool and proposes an optimized scaling strategy based on resource factors.Through the implementation of this strategy,the utilization rate of underlying resources can always be maintained in a reasonable range.In the experiment,the optimized lateral capacity expansion algorithm is validated by building a Kubernetes distributed highly availability cluster to simulate the number of load requests in the actual production environment.The experimental results show that the optimized algorithm strategy not only meets the basic functional requirements of Kubernetes,but also can advance the expansion operation and effectively reduce the application response time compared with the current algorithm strategy when the demand of approximately linear load increases.In the downsizing stage,the current resource pool is calculated periodically to release the underlying resources and reduce operating costs.