Perhaps best to start with my blog post on the subject: http://bit.ly/784kuT

yes that’s right, there will be problems when LVMs are used or VMware with VMFS, or … And impressive as it will be that the system will be able to understand Windows, Linux, Solaris, etc. FS it will still be a proprietary solution. I would prefer the extension of the SCSI protocol with commands to tell the storage array that the blocks occupied by deleted files can be released and given back into the free storage pool.

Frank

That’s how Drobo do it. However it’s tricky to make that work when people have used LVMs to combine individual LUNs together. I suspect there will be lots of caveats and restrictions.

Chris

Frank

Just to beat a dead horse….IBM XIV incorporates this ability as well. I would say at this point not having the ability is something to talk about apposed to having it, which all enterprise class arrays should have.

http://www-03.ibm.com/systems/storage/news/press/20091110.html

It’s pretty much like the HDS implementation – although that won’t work with replication (local or remote) in place.

I just checked and you need Enginuity 5874, and more importantly you need Solutions Enabler 7.1.

With that combination you can use virtual provisioning space reclamation, but there are some drawbacks.

First of all, it’s not sticky.

You issue a “symconfigure free tdev type=reclaim” against an entire thin device or use start and end cylinders and off you go. Problem is that you do this once, but after it finishes the reclaim process it won’t check again. If you have something like a thin pool that you use under vSphere for development machines it would be likely that you will see lot’s of changes on your disks, and to make good use of it you will probably want to run it more frequently.

Secondly, I can only run it against cylinders or thin devices. It would be great for me if I could run it against entire thin pools. Now I need to manually go through all disks or script my process, and with the increasing number of thin devices this is going to cost me more and more time.

What is good is that I don’t need an agent or a piece of software that will tell my array to punch holes, but it’s just not sticky so it’s a manual job.

Last point is that you need at least one thin extent, that means one consecutive part of your disk that has zeroes. Standard thin extent size is 12 tracks or 768KB. If part of the extent has data on it, it won’t be reclaimed.

Hope that helps,
Bas

Is that this is not an important event if you you are so late to the party, I mean, why advertise you came last…? Put it in the draw and when the next customer says they want/need this feature, EMC can pull it out and say they have that tick box as well – “now that you mention it, Mr Customer, yes we have that…”

But you also need to view this in relation to EMC’s own perceived value of some of these more modern technologies (i.e. specifically those brought to market by others) – in this case, Chuck’s own comments on a related topic, thin provisioning, summarises, I believe, EMC’s position:

“I think thin provisioning is not-a-good-thing at a philisophical level. It has a role, but I’d recommend using it very carefully, if at all.”

like many technologies other develop – dont believe thin provisioning is useful and one reason, I believe they use a

things

Chris