Virtualized server infrastructure has benefited from two recent advancements. First, the integration of flash is...
allowing primary storage to respond better to the random I/O nature characteristic of virtualized environments. Second, many backup applications can now recover and launch a failed VM directly off of a disk-based backup appliance. While seemingly unrelated, these two benefits are in conflict with each other.
Because of flash, users of virtual machines have become accustomed to rapid storage performance. If they are forced to run their virtualized application directly from a backup appliance, however, they will notice significant performance degradation since these systems don't use flash. Worse still, the back end of these appliances don't even use high-performance hard drives. Instead, they typically use slow, high-capacity HDDs in order to make the economics of disk backup viable.
In addition, many application vendors have started coding their applications to take advantage of the near memory-like speeds available through these resources. In short, once business applications have been accelerated through the use of SSD, there really is no going back. From a user perspective, the performance difference between going from an application being serviced by SSD resources to an application being run from conventional disk will be totally unacceptable.
As a result, data centers now have an interesting dilemma -- run VMs on storage that provides poor performance or endure slower VM recoveries back to the flash-assisted primary storage array where performance is blazing fast. The question is, what is more valuable? A VM may be able to be rapidly recovered off the backup storage system, but if performance is so slow that it's essentially unusable, then what's the point? On the other hand, uptime service-level agreements on critical business applications may not tolerate the downtime required to rebuild a failed VM back to the production storage environment.
Booting from backup cache
To solve this issue, backup solution manufacturers could integrate a small layer of flash or SSD resources into their disk-based backup platforms and reserve this area for "in-place" recovery events. That way, when a backup storage-oriented recovery is initiated, the affected VM and its business application data will operate on storage capacity more closely resembling its production configuration, resulting in more consistent application performance.
Caching backup workloads
In addition, this flash area wouldn't necessarily have to remain in a passive state, waiting for a recovery operation to commence. Instead, some of the scale-out NAS backup appliances could leverage the flash to store backup application metadata for improving backup workload tasks like data deduplication. This would make searching the backup catalog significantly faster and allow it to scale out much larger.
While installing flash in a backup storage system may seem a bit counterintuitive, the reality of a flash-enhanced primary storage tier may demand it. Otherwise, the drop-off in performance when operating from the backup storage tier may be too severe to be practical. For a relatively small investment, end users could significantly enhance the recovery capabilities of their backup systems and make their business-application recovery process a lot more seamless.
Changed block recoveries
For those environments that need a working solution today, one option is to utilize backup software technology that leverages changed block recoveries. For example, EMC's Avamar backup solution and Acronis' VM Flashback both provide the ability to perform granular recovery of virtual machine data.
In other words, instead of having to conduct a full-scale recovery of all the data from a failed VM, these products enable the backup administrator to recover only the corrupted or deleted information that was lost, providing for very rapid restores while utilizing a very small portion of network bandwidth to conduct the recovery. While there may still be some downtime on the affected application, it will be dramatically shorter than the time it would take to perform a full VM restore operation, and once the data is restored, performance will be at flash-based rather than disk backup appliance-based speeds.
The in-place recovery capabilities available in some backup storage technologies can be a valuable tool for enabling the rapid recovery of virtualized application infrastructure. The increasing use of flash storage to service VM workloads, however, dramatically reduces the ability of these disk-based systems to viably sustain VM application performance in the event of an outage.
Integrating a modest amount of flash or SSD capacity into a backup storage appliance would enable these platforms to serve a dual function as a high-performing backup repository, with the ability to serve as a failover storage system for critical VM applications. Alternatively, organizations that need a working solution today to mitigate the risks of critical application downtime may want to consider combining backup appliances with backup software solutions that leverage the speeds and efficiencies of changed block recovery technology.