One of the key capabilities that server virtualization has brought to the data center is in-place recovery: The ability to start a virtual machine (VM) directly from the backup storage target, allowing for near instant recovery of VMs without having to move data first. This feature changes how the backup appliance needs to be configured and puts increased focus on performance, especially as primary storage becomes more dependent on solid state disk (SSD).
All-flash and hybrid arrays leverage solid state storage to ease the impact of the storage I/O blender that server virtualization creates. The blender is caused by multiple VMs from a single host, all making I/O requests of the storage system at the same time and creating a very random I/O pattern.
Flash-assisted storage systems also allow VMs to host more intensive workloads that support more users and more data. In fact, applications are now being created that count on a flash-based storage environment, and users begin to expect solid state performance all the time. Executing from hard disk, especially the type common in backup appliances, may do more than just slow things down -- it may cause the application to become unusable.
[ Company consolidation is heating up the SSD market. What does this mean for enterprise IT? Read The Coming SSD War. ]
In-place recovery technology counts on disk-based backup appliances in order to serve up data. In fact, several vendors now claim that their disk appliances are better for in-place recovery because they either don't deduplicate the most recent VM image or they leverage a scale-out or grid-based architecture that can maintain an acceptable level of performance. These options are worth considering, especially if your primary storage is all-flash or flash-assisted.
I'm a fan of in-place recovery technology and backup appliances that deliver random I/O performance in addition to more traditional bandwidth performance that the backup process generates. But the time may be right for backup appliances to consider leveraging a small flash area in their storage design, to support high-performance in-place recovery.
Integration work could be done so that when an in-place recovery is performed, the VM is copied (locally) to the flash storage area. It should only take a few minutes to make this copy, and the benefits of running the VM from flash would be tremendous.
The appliance could use this flash area for other purposes as well. Storing metadata in SSD dramatically improves NAS performance, for example, and I'd expect the same from a backup appliance, especially ones that are scale-out and have deduplication to manage.
It may seem odd to consider solid state storage in a backup device, but 15 years ago the notion of using disk instead of tape for backups seemed equally odd. Today it is commonplace, and flash is becoming more affordable every day. The increased usefulness of the backup appliance would more than compensate for the higher price.