Shared-nothing live migration allows running virtual machines (VMs) to move between two separate standalone servers, without any clustering or shared storage between them. It can use a connection as small as Gigabit Ethernet (GbE).
The feature works by moving the VM storage first, including snapshots, the virtual hard disk and configuration metadata, over the Ethernet connection between servers. Once that’s done, the running state of the VM, including the smallest possible subset of in-use memory, is then moved. (This subset of memory, as well as the memory available on the destination server, is determined through a “pre-flight check,” another new feature previewed for Hyper-V 3.0). Finally, once the second VM is up and running, the VM state is deleted from the source server.
A plus for planned maintenance
The ability to live migrate VMs without shared storage is particularly appetizing in shops where shared storage is impractical from a cost standpoint, but that still want to use live migration for planned downtime.
“I have a smaller environment where I don’t have shared storage… I would definitely use something like this,” said Femi Adegoke, IT Director at the West Gastroenterology Medical Group in Los Angeles, Calif.
Adegoke, who ran through a live demo of shared-nothing live migration at Microsoft’s Build conference earlier this month, said he is considering it for use in branch offices where direct-attached storage is the only option.
Even where virtualization clusters already exist, shared-nothing live migration could provide Hyper-V administrators with additional flexibility when performing planned maintenance, by allowing VMs to move in and out of a cluster as needed.
“Before, it had to be overnight on a weekend to be able to shut stuff down to do that work,” said Joe Henrich, assistant vice president of technology at a regional bank in the Midwest.
With shared-nothing live migration, he can move the VMs to another machine outside the cluster for a short time while performing maintenance on the main clustered environment.
Conversely, “the same thing [applies] when I decide a VM needs to go into the cluster for high availability,” Henrich added. “It’s more about giving me the ability to do things during business hours as opposed to after business hours.”
Not a replacement for live migration with high availability
While shared-nothing live migration has its use cases, it won’t completely replace Microsoft’s existing live migration with high availability, since shared-nothing live migration is for planned moves only. Unplanned failover of either storage or VM state requires a “hinge point” on one side or the other.
There are also limits to the scalability and performance of migrations users can expect over GbE links. A terabyte VM, for example, would be impractical to move.
As such, most users will shy away from moving large VMs without shared infrastructure.
“If you’ve got a great big box, I probably wouldn’t do it live… in the middle of the day,” said William Bressette, a network architect at Horn IT Solutions. “That’s just kind of a best practice thing.”
Beth Pariseau is a senior news writer for SearchServerVirtualization.com. Write to her at email@example.com.