一种基于Docker容器的集群分段伸缩方法

摘要

A scaling method of cluster is proposed based on Docker container technology to solve the problems of present scaling method, such as slow response and heavy cost.It uses auto-regressive model to predict future workload while monitoring the real-time workload, and calculates the need of resource with one queue theory model.This method first relocates resources by itself, then it chooses the Docker scaling or VM scaling depending on the workload change rate until the response time is within an acceptable range of users.The experimental results show that this method can provide faster scaling response and less cost when dealing with fluctuating workloads.%针对现有的虚拟机集群伸缩方法响应慢、开销大的问题,提出一种基于Docker容器技术的虚拟机集群伸缩方法.在检测实时工作负载同时通过自回归模型对未来工作负载进行预测,最后使用排队论模型计算所需伸缩量.首先从虚拟机自身进行资源重分配,然后依据工作负载变化率选择Docker容器级别的伸缩或虚拟机级别的伸缩,直到请求响应时间在用户可接受范围之内为止.实验结果证明,在面对不同的工作负载变化情况时,该方法可以提供更快的响应速度和更低的开销.