I'm looking at my Symmetrix array, and over 3 years or so of 'adds and deletes' of LUNs, we've ended up with quite a lot of 'gaps' showing up on our disks.
As these things turn out, these gaps seem to have ended up inconveniently sized, such that of our theoretical available capacity, not all that much is really usable.
So I was wondering - has anyone tried to defragment the symmetrix disks, by relocating the hypers?
I'm... well, sort of thinking it'll be difficult, because a symconfigure create won't let me control where the hypers go precisely. But they do, at least, use a predicatable pattern, and therefore it's possible to predict an outcome.
So I could:
Identify the 'gaps' I want to consolidate, and the LUNs that come 'after' said gaps.
Create new hypers - of the right size/geometry - elsewhere - either in the same disk group, or ideally (for this purpose at least - I recognise it may not be optimal elsewhere) a separate DG.
Use Symmetrix optimizer to manually swap - such that your 'real' data is moved onto the temporary devices.
delete the devices that have been swapped onto your existing disk, and re-create them, such that they'll be right at the end of the existing data.
And then use symmetrix optimizer to swap back.
This might need to swap off a bunch of hypers, delete and recreate them, but if you're lucky/smart about it, you may well be able to get away with just a little bit of re-arranging - that gap is _probably_ of a standard-ish size, that's been left because you deleted a lun, so you can probabably 'fill' it, and reshuffle accordingly.
Now, I thought I'd post here as a sanity check - does this sound feasible?
Does functionality like this exist already/is there a better way of doing it?
Please note - I'm talking about fragmentation of the hypers on the symmetrix disk groups, not the host filesystem - that we've got host-side tools for.
If there's a bunch of holes, I would recommend getting your local EMC CE to build you a bin file that reclaims that unused disk storage and makes it available in predetermined sym vol sizes or metas. This would be much quicker I believe.
It probably would, yes. I'm just sort of thinking that getting EMC in to do it is sort of analagous to the 'manual defrag' approach, where a scripted solution could - at least in theory - be used to keep on top of the problem.
the amount of effort trying to explain to a CE what you are trying to do isn't worth it I don't think
only other thing to consider is how far away your next symmetrix refresh is, you could sort it out then
> the amount of effort trying to explain to a CE what you are trying to do isn't worth it I don't think
I think that depends on your CE. If you say things like this about your CE, I suggest you ask for another one. I'm always very happy with my regular CE !
Our CE is good, and I'd have no doubt they could and would do this.
I'm just ... well, looking for an excuse to do some perl hackery, really .