Not quite hyperconverged Backup

Why do backup products not come integrated into the storage that they require? Pretty much every backup product that you deploy wants a disk target to write to. That disk target has an x86 CPU in it and the backup software could run on that CPU. This is one thing Rubrik have done. Their product is a hybrid storage array that also runs their backup software. This backup array is a deduplicated scale out storage appliance, using SuperMicro Twin hardware it puts a four node cluster in 2U. A basic 3 node configuration should backup around 200 VMs, about 100TB of backup capacity. For scale more of these units can be added, apparently without practical limits to how many units.

Normally a backup target would not need tiered storage and would have a far higher disk to compute ratio. This points to Rubrik’s expectation that customers will also use their backups for a tier two primary storage array. This will allow full performance instant VM recovery as well as using the Rubrik array to run test and development copies of production VMs.
Instant VM recovery is a great feature, first seen in Veeam, that allows a VM to be recovered simply by booting a copy from the backup store. The backup store presents an NFS datastore to the ESXi servers. This datastore contains a copy of the VM that is reconstituted from the deduplicated backup store. The VM is registered on ESXi and powered on. This allows a VM to return to service in minutes rather than the hours that might be required to restore the whole VM to primary storage. The performance of the recovered VM is dependent on the capabilities of the backup array. With Rubrik the backup array should deliver excellent performance, reducing the need to immediately relocate the VM to the original primary array.
The other big thing that Rubrik emphasized is the indexing and search. The VM disks are catalogued, allowing search for individual files within VMs and restore of individual files. This will greatly aid backup administrators tasked with restoring files that users accidentally deleted. As we know the user seldom knows what the file was called or where it was saved.
Overall I like the Rubrik storey. It sounds like they have a solid background in storage and getting value from stored data. The architecture looks a bit like a hyper-converged solution, but does not provide VM execution. This means Rubrik has scale-out for capacity and performance. There shouldn’t be any hardware bottlenecks as you scale to huge capacity. I would like to see them add a backup agent on the ESXi servers. Currently backups are completed by NBD, which can be very slow. Having a virtual appliance on the ESXi servers would allow for the faster hot-add backup methodology.

Disclosure: I learned about Rubrik at Virtualization Field Day 5. All travel and incidental expenses for attending Virtualization Field Day 5 were paid for by Gestalt IT.  This was the only compensation provided, and it did not influence the content of this post.

© 2015, Alastair. All rights reserved.