Neural Network Based Regression Model for Virtual Machines Migration Method Selection

Abstract

Live virtual machines migration has been one of the main strategies in cloud systems management to efficiently utilize data centers resources, reduce power consumption and unutilized resources in data centers, as well as providing the least interruption to customers in the events of migrating virtual machines or their data between different hosts in the same or different data centers. Many migration methods, with different characteristics, have been proposed to migrate virtual machines. Pre-copy and Post-copy are the main classical migration methods that transfer VMs and their memory between different hosts. The main VMs migration performance metrics include the total migration time, downtime, and amount of transmitted data. Machine learning algorithms have been widely used to make systems intelligent. Neural Network is one of the key machine learning algorithms used for classification and regression. In this paper, we propose neural network-based adaptive models that predict the key performance metrics of VM migration for Pre-copy and Post-copy methods, and for different application workload types running over the virtual machine. Based on the prediction, one of the two migration methods can be selected to migrate a specific virtual machine.