Analysis and Comparison of Live Virtual Machine Migration Methods

Abstract

With the huge development of servers' virtualization technology, the importance of live virtual machines (VM) migration has increased in order to improve hardware utilization, increase the power efficiency of data centers, and to decrease the effect of data centers failures and service downtime during the maintenance period. Live VM migration is the process of copying the CPU state and the memory and disk states of a VM from a hosting physical server to another destination server at the same data center, or to another data center connected to the hosting data center over LAN or WAN networks, with the least service downtime, in order not to affect the Quality of Experience (QoE) of virtualization service users. After the VM is transferred to the destination host and before it resumes running there, network traffic should be redirected to the new VM's location. In this paper, we analyze and compare the three migration methods: the Pre-copy, the Post-copy, and the Hybrid-copy method. We explain the mathematical model of the Pre-copy method as available in the literature. We present our mathematical models for Post-copy and Hybrid-copy methods for the migration downtime, total migration time, and total transferred data during the migration process. We implement the three models using MATLAB and evaluate the performance of these migration methods for different experimental parameters. Based on the performance we propose which method is suitable for live migration of virtual machines.