一种改进的Kubernetes弹性伸缩策略  

An Improved Kubernetes Elastic Scaling Strategy

作　　者：沐磊 李洪赭 李赛飞[1] MU Lei;LI Hongzhe;LI Saifei(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756)

机构地区：[1]西南交通大学信息科学与技术学院,成都611756

出　　处：《计算机与数字工程》2022年第2期327-331,372,共6页Computer & Digital Engineering

基　　金：四川省重大科技专项课题(编号:2018GZDX0005,2019YFG0399,2019ZDZX0007);中央高校基本科研业务费专项(编号:2682019CX63)资助。

摘　　要：Kubernetes是目前主流的容器云编排和管理系统,其内置的伸缩策略是通过监测衡量指标并与阈值比较计算,从而实现伸缩的功能。该策略主要存在单一衡量指标和响应延迟问题:单一指标在衡量多种资源消耗的复杂应用时存在明显缺陷;响应延迟问题会造成应用在一段时间内的服务质量无法得到保障。针对上述问题,提出了一种改进的Kubernetes弹性伸缩策略,该策略对应用涉及的多种资源进行计算,得出综合负载率作为衡量应用伸缩的指标;使用ARIMA-Kalman预测模型对综合负载进行预测,从而实现预测式伸缩,提高了应用在面对突发流量的应对能力。实验结果表明,该策略能够较好地衡量应用的整体负载水平,并能对应用负载进行准确的预测,解决了响应延迟的问题。Kubernetes is currently the mainstream container cloud orchestration and management system. Its built-in scaling strategy is to achieve scaling by monitoring metrics and comparing with thresholds. The strategy mainly has a single measurement index and response delay,a single indicator has obvious shortcomings when measuring complex applications that consume multiple resources. The response delay will cause the application’s service quality to be unable to be guaranteed for a period of time. In view of the above problems,an improved Kubernetes elastic scaling strategy is proposed. This strategy calculates multiple resources involved in the application and obtains the comprehensive load rate as an index to measure the application scalability. The ARIMA-Kalman prediction model is used to predict the comprehensive load,so as to achieve predictive scaling and improve the application’s ability to cope with sudden traffic. Experimental results show that the strategy can well measure the overall load level of the application,and can accurately predict the application load,and the problem of response delay is solved.

关 键 词：Kubernetes 弹性伸缩 综合负载 ARIMA模型 卡尔曼滤波 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]