Wednesday, December 16, 2009

To enable direct recovery of virtual machines from backup first requires a disk based target that can present itself as an NFS mount point. Ideally, this disk system would also have deduplication, like Data Domain's DD family of products. These systems are ideal for virtual environments where there’s both highly redundant data between backups, and highly redundant data between virtual machines. Deduplication makes replication of backup data to a remote site a reality as well.

The second requirement for direct recovery is a virtual environment that supports the booting of virtual images from NFS partitions. Booting from NFS brings many advantages to the IT administrators over traditional block fibre channel solutions. It’s important to understand, however, that for the purposes of recovery it’s not a requirement that booting from NFS already be in place, just that it can be done when a recovery is needed.

The final requirement is the ability to capture copies of the virtual machines in their native state. In other words, the classic backup application will not work here, as most backup applications store data in some sort of backup format that’s only accessible through the backup application. Direct recovery needs the copies of the virtual machines to be in their native format so that the virtualization host can directly access them.

The simplest way to manage this is to leverage the snapshot capability built into most virtual server environments. These snapshots suspend the virtual server for a few seconds so that the cache can be flushed and an image of the virtual machine can be presented in a static state. Like other snapshot technologies, this virtual machine image only consumes disk space when an active version of the VM is updated with new data. Also, the snapshot is totally dependent on the active copy of data remaining up. Failure in the hard drive system that the active data resides on would lead to failure on all the snapshot copies. Therefor, creating a copy of this snapshot to a secondary disk system is an ideal way to free up storage resources and protect against primary hard drive failure. Again, the ideal target for this copy is a disk based deduplication system since it will enable native access to the data and store that data efficiently.

Once these components are in place it becomes a relatively simple matter to script the taking of a snapshot of the virtual machines and the creation of its copy on the deduplication system. If there’s a second deduplication system in place, replicating that data to a disaster recovery site is a simple matter. The only real bandwidth concern is moving the entire virtual image across the local network. For most environments the network infrastructure can handle this sort of transport. If there’s an issue, utilities are available that will copy virtual machine image data at a block-level across the internal network, to the target device.

With everything in place, testing direct recovery is a matter of mounting and booting the VM via NFS and bringing it to an active state. These disk based systems, especially ones with deduplication enabled, are fast but not typically designed to be production storage. That said, with the direct recovery complete, the VM and its associated application is up and accessible by users. With these VMs come online tools from the virtual server environment provider which can be used to move the storage to a production system. For example, VMware's Storage vMotion can move the storage for a VM from the deduplicated backup device to primary storage without the user experiencing a service interruption.

Direct recovery provides a number of benefits. First, leveraging deduplication systems, provides a very cost effective way to protect the virtual environment. Typically, deduplication rates in the virtual environment are often 20X more efficient than in a non-deduplicated environment. Deduplication also enables the cost effective replication of data to a disaster recovery site. Secondly, direct recovery enables NFS based deduplication systems to provide direct access to data from the virtual environment. This means that virtual machines can be available to applications and users in seconds in the event of a failure, since there’s no delay waiting for backup data to be translated from a proprietary backup format and copied across the network. Finally, as the VMs are restored to operation they can be copied in the background to more traditional production storage without interruption to the users.

George Crump, Senior Analyst