Abstract
Live migration is one of the most useful features provided by todays' virtual machine monitors (VMM). It enables seamless hardware upgrades, provides fault tolerance, achieves load balancing and saves power through server consolidation. These features can also be beneficial in HPC environments. This paper presents a comprehensive study of the migration facility of the Xen VMM, specifically targeting HPC applications. We analyze the effects of live and non-live migration techniques on HPC application wall times. A detailed relationship of the migration routine with memory modification, communication intensitivity and CPU contention between guest VMs and the host VMM is presented. We propose a simple optimization for the live migration feature. Our optimization is able to reduce the total number of memory pages transferred during the migration by up to 500% and results show an average of 50% improvement over the default Xen migration routine on the traditional gigabit Ethernet infrastructure. We also demonstrate that live migration of virtual machines in an HPC environment can be used to improve application wall times.