There are problems with any kind of automated storage tiering but does that mean it's not worth doing?
I think it is worth looking at the kinds of problems that storage tiering is intended to fix and especially storage tiering of the automated kind. I guess at the simplest level, it is about getting your data onto the right type of storage and ensuring that you get the best use of the various resources that you have. However defining what is the right type of storage is not easy and it also may change over the lifetime of the data.
Automated storage tiering can address a number of objectives and we might have to also be pragmatic about how it will be misused as a substitute to what we may feel is best practise.
Automated storage tiering may allow us to make much more efficient use of an expensive resource or at least what is perceived to be an expensive resource.
SSDs are currently expensive if you look at them purely as places to store data at rest or data that has less than demanding access requirements but if we consider them as the place where our most I/O intensive workloads live; they become more economic.
SATA disks are cheap if you look at them as purely places to store data pretty much at rest but it we consider them as places for our most I/O intensive workloads to live; they become less economic, especially if you take into account environmental considerations.
FC disks are at the mid-point and cannot yet be discounted from the equation as much as people might like to. This is going to change and it is this mid-tier which will eventually disappear.
So automated storage tiering should allow you to use disk more efficiently. And it might bring some unexpected benefits; for example, how many people have legacy applications which have databases which simply grow forever?
Be honest, lots of us have; much as I would like to pretend differently, it is a fact of life for many people that one of the first non-functional requirements which get sacrificed in any development is the ability to get data out into an archive.
And many of us have legacy applications which are never going to be fixed and will simply grow forever; at present, they sit on FC disk which more or less does the job but it is not especially economical on a pure IOPs measure or a pure capacity measure and it gets less so. But to take advantage of a SSD/SATA mixed environment would require a massive amount of rework and continuing maintenance. Automation is the key to this, especially automation which allows us to optimise at the sub-file and sub-LUN basis.
The right thing to do would be insist that archiving gets integrated to the applications but the pragmatic thing may be to utilise storage tiering to do this.
We all agree that archiving is a good thing but so is exercise.....and if they invented a pill to stop people getting fat even fewer of us would exercise despite the additional benefits that exercise brings. But just curbing obesity would be considered a major advance; we need to at least consider automated storage tiering in the same light...
And for all you BURA specialists out there; I feel your pain but pragmatic short-term fixes win very often in this space.
p.s brownie points for anyone who gets the reference in the title....
Martin, I didn't get the reference, but I assume it has something to do with American Express? It's primarily a charge card and not a credit card. That's as much as I could do with it.
The aging databases you mention typically don't need SSDs today - right? In a perfect world there would be integrated archiving and SATA-SSD tiering, but an imperfect (affordable) world there is something else. Why not create two tiers from FC and SATA, with most of the new storage being SATA drives? It's not without work, but there are automated tools from some vendors - for instance, 3PAR has Dynamic and Adaptive Optimization that can do most of this work.
Posted by: marc farley | March 10, 2010 at 10:28 PM
Pragmatism FTW! Could be best summed up as I often respond to my children:
"In a perfect world, you'd have unlimited funds, free time and someone servicing your every whim. Since this isn't a perfect world, go take out the trash until then please."
Title is a spin on Visa's slogan. What are brownie points good for... in a perfect world?
Posted by: John Dias | March 10, 2010 at 11:56 PM
Marc, actually many of the aging databases I am thinking about could really do with SSDs. They are currently spread over hundreds of FC drives for pure performance reasons; SSDs would be ideal for them but they also continue to grow and we know there is aged data which could be moved to a slower tier but the application architecture makes it very hard to do so.
And the reference, is actually a long convoluted reference based on Automatic and I wouldn't expect any American to get it. In fact, unless you follow British guitar-based bands, you'd never get it.
There is a British band called 'The Automatic'; somewhat derivative indie-guitar pop but still fairly listenable; 'Monster' was their big single. Their debut album was called 'Not Accepted Anywhere'...which is obviously a play on the 'Accepted Everywhere' slogan.
Posted by: Martin G | March 11, 2010 at 08:45 AM
I looked them up and apparently they are known as The Automatic Automatic in North America something about legal issues surrounding names. Anyway I'm listening to Monster now - God I love the Internet!
Posted by: marc farley | March 11, 2010 at 10:03 PM