Analysis and prediction of virtual machine boot time on virtualized computing environments

Abstract

Starting a virtual machine (VM) is a common operation in cloud computing platforms. In order to achieve better management of resource provisioning, a cloud platform needs to accurately estimate the VM boot time. In this paper, we have conducted several experiments to analyze the factors that could affect VM boot time in a computer cluster with shared storage. We also implemented four models for VM boot time prediction and evaluated the performance of the four models based on the datasets of four hosts and seven hosts in our environment, where the four models are the rule-based model, the regression tree model, the random forest regression model, and the linear regression model. According to our analysis, we found that host capability and maximal network bandwidth are two main factors that can influence VM boot time. We also found that VM boot time becomes harder to predict when booting VMs at different hosts concurrently due to competition between hosts to obtain resources. According to the experimental results, the proposed random forest regression is the best model for VM boot time prediction with an average accuracy of 94.76 $$\%$$ and 96.59 $$\%$$ in predicting VM boot time in two clusters with four and seven compute hosts, respectively.