面向微服务架构的容器级弹性资源供给方法

发布时间：2018-02-20 02:11

本文关键词： 容器虚拟化模糊自适应式卡尔曼滤波弹性资源供给微服务架构突发性负载　出处：《计算机研究与发展》2017年03期 　论文类型：期刊论文

【摘要】：容器作为物理资源的逻辑抽象,具有资源占用少、资源供给快等特点,适合工作负载突变的互联网应用模式,特别是面向微服务架构的新型服务范型.已有工作受限于物理机和虚拟化环境,或资源难以弹性供给或资源供给时效性较差,难以应对负载突变(flash-crowds)场景.针对此问题提出了一种服务质量(quality of service,QoS)敏感的、基于前馈的容器资源弹性供给方法,该方法采用排队论刻画工作负载、资源利用率和响应时间的关联关系,构建应用性能模型.其中,响应时间采用模糊自适应卡尔曼滤波进行预测(前馈控制器),预测结果违背QoS是触发资源弹性供给的依据.基于CloudStone基准的实验结果显示,前馈控制器具有快速收敛的特点,对响应时间的预测误差小于10%.在flash-crowds场景下,相对于已有方法可有效保障应用的QoS.
[Abstract]:As the logical abstraction of physical resources, the container has the characteristics of less resource occupation and fast resource supply, which is suitable for the Internet application mode with sudden workloads. In particular, new service paradigms for micro-service architectures have been developed that are limited by physical machines and virtualized environments, or where resources are difficult to provide flexibly or which are not available on time. It is difficult to cope with the sudden change of load flash-crowds. in order to solve this problem, a sensitive and feedforward based flexible supply method for container resources is proposed, in which queue theory is used to describe the workload. The relationship between resource utilization and response time is used to build an application performance model. The response time is predicted by fuzzy adaptive Kalman filter (feedforward controller). The result of prediction is contrary to QoS, which is the basis of triggering elastic supply of resources. The experimental results based on CloudStone benchmark show that the feedforward controller has the characteristics of fast convergence. The prediction error of response time is less than 10. In the flash-crowds scenario, compared with the existing methods, the QoS can be effectively guaranteed.
【作者单位】：中国科学院软件研究所软件工程技术中心;计算机科学国家重点实验室(中国科学院软件研究所);中国科学院大学;
【基金】：国家自然科学基金项目(61472407,61363003)~~
【分类号】：TP393.09

【相似文献】