There is this great story that describes how the railroad gauge of 4 feet and 8.5 inches derives from the Imperial Roman War Chariot. The point behind that story is that we continue to do things long after the reason for doing them has evaporated.
It turns out that tape is playing the same role in the data center.
So it turns out that backup is the only ubiquitous data protection policy. Regardless of how small you are, protecting data from operational errors (human frailty and software frailty) is something you do. The first time you accidentally delete an important file and the first time an upgrade fails is the moment a backup procedure gets created. And that tends to happen within a few months of acquiring a computer system.
Because backup is so ubiquitous it's very highly automated. There is an awe inspiring amount of software developed and sold for backup.
And once you have a backup policy, the next thing you start to worry about is disaster recovery. Namely, how do you recover all of the data after a system or site goes belly up. And the obvious thing to do is leverage the existing backup infrastructure to get the data back after a disaster. And then once you've got disaster recovery covered, the next thing to worry about is long term archival. And it's logical to just leverage the same backup infrastructure, just hold onto the backups for longer...
So far so good.
And back in the Ur-days when tape was a lot cheaper than disk, the only way to economically do backup was to use tape.
Okay so back when you just did backup and restore, you probably chose to do a single full backup followed by some large number of incremental backups. You probably held onto the tapes for a modest amount of time and you had a small number of tapes. Life was good.
But then you wanted to leverage Disaster Recovery so you had to start taking more full backups more frequently. Why? Because doing full recovery from a set of incremental backups is simply not practical. And all of a sudden you had more tapes, and things started to take more time, and you needed bigger pipes.
And then you wanted to do long term archival, and all of a sudden you had even more tapes staying around for longer.And now you had a whole set of policies that you needed to put into place to recycle tapes.
So now what?
Well now we have a set of processes and procedures that exist because that's the way tape made things exist. We have servers that have enough IO bandwidth to be able to do full backups on a regular basis. We have network infrastructures that only exist to sustain the vast amount of data that needs to be moved for backup. We've created and sized an entire infrastructure predicated on moving all of the data very frequently because that's how tape made us do it...
Okay...
But here's where it gets insidious, we know look at the mess that tape has created, and instead of asking the question:
Is a data protection infrastructure predicated on creating whole copies on a regular basis flawed?
We ask the question:
How can I make creating and storing full copies more efficient?
Rather than use new technologies that make backup and restore efficient, disaster recovery efficient and archive efficient and effective, we continue to rely on creating vast amounts of redundant copies, keeping those copies around for a while and praying that deduplication will save our skins.
And when folks argue that deduplication really should be done on the primary, and that we should look at ensuring that as little data moves we act surprised. Data management is about managing thousands of whole copies, that's the way it's always been ...
At some point, 50 years from now people will ask why do we expend so much energy eliminating redundant data when we could have avoided creating the data in the first place. And someone will write an article about the role of tape and whole images and people will laugh...
Comments