Migrating a virtual machine from one host to another without using shared storage is the current virtualization party trick. Microsoft introduced the shared nothing feature with Hyper-V 3.0, and VMware added it to vSphere 5.1.
Even though the shared-nothing feature migrates virtual machines (VMs) without shared disks, it still uses a shared network. This migration type allows the hypervisor to move files that comprise the VM from one location to another. Because the files must be moved to a different physical storage device, shared-storage migration takes longer than moving memory content and CPU state, as is required for migrations with a shared disk.
Because these migrations take longer and put a larger strain on the network, they aren't automated. Instead, most admins will start the process manually as part of a maintenance activity or other operational change. So, when is this extra migration time worthwhile?
More on live migration
Shared-nothing live migration in Hyper-V 3.0 a benefit for small shops
Troubleshooting storage and network configurations for live migration
The ins and outs of RHEV live migration
Some applications have a scale-out approach to availability, meaning multiple VMs work together to host the application. The application doesn't require specific VMs to be running; it only requires that enough VMs run to process the current workload, such as is the case with a Web server farm or terminal server farm.
Because dividing the application among multiple VMs provides availability, there is no need to have availability at the hypervisor. VMs can be placed on local storage in the host, saving space on the storage area network. Of course, you need to spread VMs across multiple hosts, so one host failure doesn't greatly affect the application.
You can encounter a problem, however, when you have to perform maintenance on a host. Migrating VMs to a different host would facilitate maintenance, but without shared storage, taking a host out for maintenance would require you shut down all VMs on its local storage.
It's not all about local storage
The first thing people think of when they hear migration without a shared disk is that it's all about VMs that reside on local disks. This may not be the case for enterprise customers that often have multiple clusters of hosts.
Each host is connected to others within its cluster through shared storage, but often storage is not shared between clusters. One cluster might be for development and preproduction while another is for production. The ability to promote a VM from preproduction to production without an outage or shared disk between the clusters can be very beneficial. This sort of flexibility is what makes virtualization so attractive.
Another place where VMs might migrate from host to host without shared storage is migration between hosts at different physical sites. The allowed separation between the sites won't be far -- a few dozen miles at the most. But this distance could be far enough to avoid downtime during a disaster. Relocating a VM from one side of the city or state to another may be enough to avoid the power outage, fire, flood or other minor disaster that would otherwise involve a service outage.
Yet another possible use for shared nothing live migration is to migrate a workload into or out of a cloud provider, within the same city or state, making cloud bursting more practical. The ability to move an enterprise application from an on-premises cloud to a public cloud provider without downtime will be a critical part of hybrid cloud operations.
VM migration without shared storage is a reality that enables greater VM mobility than traditional shared-disk migration. It's another game-changing feature of virtualization.
This was first published in October 2012