Storage vMotion of a Virtualized SQL Server Database

This paper provides a deep look at the performance implications of the live storage migration of a VM running a large SQL Server database workload. Performance of the workload when migrating different components of the database (data, index, and log) was measured and compared against the steady state performance (when there was no migration). Experiments discussed in this paper also show the impact of I/O characteristics of the database components on the migration time of the virtual disk containing them. Based on the knowledge gained from the experiments, a few best practices are suggested for the live migration of the storage of the virtual machines (VMs) with minimal impact to the performance of applications running in them.

Live storage migration is the missing piece in liberating VMs and their associated files completely from the physical hardware on which they reside. Predictable migration times—with minimal impact on the performance of the application accessing the virtual disk that is migrated—are expected from the vSphere’s storage vMotion feature. This paper offers a deeper look at the interaction of svMotion with a large, active SQL database workload. The study includes application behavior when migrating individual virtual disks used by the database and the impact application I/O traffic had on the svMotion of a particular virtual disk. The study showed consistent and predictable disk migration time that largely depended on the capabilities of the source and the destination arrays. svMotion increased the CPU consumption of the VM running the test workload from 5% to 22% depending on the load conditions. The I/O patterns of the SQL database workload had noticeable impact on svMotion throughput (and the disk migration time).

 Storage vMotion of a Virtualized SQL Server Database