"There are two kinds of disks in the world -- those that have failed and those that are going to. So we have to be concerned with media failures. How do we normally prepare ourselves against the possibility of media failure? We use RAID," said Jon Toigo, founder and CEO Toigo Partners International, who noted that problems with interfering with RAID setups led to changes in how RAID is implemented.
For example, after RAID 1 fell out of favor, users turned to RAID 5, which appeared to be a better approach from parity striping. In the past, he said that's because if a disk failed, the data can be rebuilt from parity information stored in all the drives in the set.
However, RAID 5 is still susceptible to data loss if more than one drive fails at the same time.
"By the time you respond to the placement of a drive, a drive light goes on, and now your RAID set is operating on just its primary spindles and one of its disks is dead. By the time you address that problem and replace that drive, you're going to have a second or third failure," said Toigo. "That's happening with greater and greater frequency. When that happens, RAID 5 doesn't get you where you need to go."
"Classically, what we've done is tape …. You install the tape appliance, the tape library, and you make copies of your production data at routine intervals to tape. There are a lot of variations on a theme…. We went from classic tape to tape with backup server," he said.
"Basically, what we were going to do was address the problem of how slow backup is and how all the backup processes -- the superstream -- fall apart when the smaller tape jobs complete and we're not driving the tape library's rated speed anymore. Instead, what we do to normalize the exchange of data between the central repository and the tape library is install a server that is dedicated just to backup. And we make all of our writes of data initially to that backup server and we normalize the speed and feed of the backup job to the back-end tape library. It's sort of like a primordial VTL."
Toigo went on to say that "[VTLs were] basically building on the concept of that backup server and creating a disk appliance. And we had two reasons for doing it: not only to normalize the exchange of data back and forth between the tape library and the disk repository, but also to create a location where we could temporarily store data so we could do individual file restores at the drop of a hat. So you can restore 30 days' [worth] of data in a virtual tape library. And if an individual file becomes corrupt -- which is more often the case than a total facility meltdown -- you can rapidly restore without having to reload an entire tape backup set and find the file you're looking for. That was considered to be a real innovation in its day."