The first part of my discussion centered on the core issue in enterprise storage: Despite the cost of capacity dropping rapidly, storage really hasn’t gotten any cheaper.
Thin provisioning doesn’t change the actual cost of capacity. Instead, thin provisioning attacks the overhead of inefficient provisioning. Not all overhead is inefficiency, and not all that is can be tackled with thin provisioning. But some of it can, and this is a lot more of the cost than can be tackled by moving to SATA, for example.
Let’s look at where the costs are. One thing that we’re not going to be able to affect is the cost of data centers and networks. You’re not going to cut out the cost of operations and administration. Storage people are generally overworked already, so any time you save in terms of allocation is going to be spent on some other nightmare in storage. There are plenty of them, like backup, archiving, compliance, all those kinds of stuff. We’ve got a lot of work to do.
So, really the only cost component that you can attack are the cost of hardware and software (which is relatively small and can’t ever be eliminated completely) and the inefficiency due to utilization.
So inefficient utilization is the right target. Why is storage allocation inefficient?
- Let’s say I’ve got some mythical application that uses a stable amount of storage. That’s the blue box on the left in this diagram.
- The application guy knows that it takes months for the storage group to respond to capacity requests. And they know that they have to take their system down to add it, so they’ll have to schedule and wait for an outage window – all this kind of stuff. So they’re just going to go ahead and order more storage than they really need. That’s the brown box.
- Now I’m the storage guy, and I’m going to buy a huge pile of storage for this application. But I am doing the same thing the application guy is! I want to make sure that I have enough to provision. I also have to buy extra because I must have data protection. I have to have all sorts of other things out there to meet these demands. So I buy too much. That’s the tall dark blue box.
- They can’t actually use all of it, though, because of data protection. The usable storage is the light blue box, and it assumes a pretty darn efficient system!
- I only allocate some of the capacity right now. I give them what they request – that’s the light brown box.
- They really aren’t even going to use all of it right now. That’s the dark blue box on the right.
Look at the big picture and you see that we’ve got a small amount of storage used and a lot more storage purchased. We can’t attack every part of the problem, but we can attack some of this overallocation with thin provisioning. That’s our next topic.