Adaptive Downtime for Live Migration of Virtual Machines

摘要

Live migration of virtual machines is a pivotal mechanism to offer consolidation services in virtualized and cloud computing environments. In general, the standard live migration process favors short downtime periods, to avoid affecting on-line services provided by virtual machines. In particular, live migration can only be applied in scenarios where the dirty page rate is low compared to the communication network bandwidth. Furthermore, while setting a short downtime is convenient for on-line services, this may not be necessary, e.g., For long time-consuming high performance applications as, in this sort of computations, a relative large of downtime period, (say, in the order of minutes) due to a live migration, only introduces a small overhead on the total execution time. In this paper we enhance the live migration process in a variety of scenarios where the standard live migration configuration does not permit live migration. For these situations, we propose, analyse, and validate a modification of the KVM hyper visor that accommodates live migration with short downtime, independently of the type of application and services in execution on the virtual machine.