Enterprise Computing: So EMC, Where's Your Thin Persistence?
Enterprise Computing — By Chris Evans on October 13, 2009 at 6:34 AMFollowing on from yesterday’s announcement by 3Par, I’m wondering just where EMC is in the thin reclamation market. HDS have ZPR, which although working as a background collection process, does reclaim zeroed out blocks of data. I believe SVC has similar functionality too.
We all know that over time, Thin volumes transition back to Thick volumes as data is created and deleted. I suspect virtualisation platforms such as VMware will make that even worse. So where are EMC in this market? Bearing in mind FAST has been announced well ahead of potential availability, I can only assume EMC don’t have anything on the horizon. Make sure you consider this when you’re weighing up the benefits of a thin strategy.
Oh and its nothing to do with block size – after all, 3Par have the smallest block allocation and still see the need to provide persistent thin technologies.
If you’ve not read it, here’s my recent summary on thin reclaim.
Tags: 3par, EMC, HDS, InServ, thin provisioning, ZPR
(No Ratings Yet)
Loading ...
Tweet This
Digg This
Save to delicious
Stumble it
6 Comments
There’s a way… but it’s not a real “fun” way. If you are a powerpath user, you can use it’s migration enabler to do thick to thin conversions live (at least for Linux & Windows). Here’s where is stops being fun: you present the host new target luns, then you use powermig to sync the source to the target luns. It basically looks for 256k of sequential zeros, and throws those away, if you have a 255k of zero’s and 1k of data it will write everything (the 256k was from memory so it might be off). So for me if I really want to reclaim as much as possible: defrag first (if on windows) then create a large file with nothing but zeros in it, then delete the file so I have the best chance of contiguous zero’s (which is true for any array’s zpr)
This is all done online so I don’t have to do a whole lot of application/customer negotiation for time, but it’s a big pain in the ass if you ask me compared to other vendors of just running it as a low-priority service in the array (but at least the option’s there).
I think you’re right – not a fun way? Surprised EMC haven’t responded…
Chris
Chris, you are correct SVC 5 brings Zero Impact Zero Detection function to our “Space-efficient” volumes, allowing us to also go from “thick to thin” as well as stay thin by inline detecting zeros.
What is interesting is that 3Par think they need a custom ASIC to do this – when SSE 64bit macros in commodity Intel CPUs provide an almost zero impact mechanism to implement said function.
Null/0 page reclamation has a rather limited set of use cases. Probably the most effective use being to reclaim null/o pages after a thik to thin migration assuming you don’t have access to something like Symantec SmartMove.
After that 0 page reclamation is much less useful because filesystems do not typically write lots of blocks full of nulls. When a filesystem deletes data it typically does not write nulls to the blocks that are no longer used, in stead it de-references the blocks, because these blocks do not contain nulls 0 page cannot reclaim them.
This is one reason why data recovery programs can recover deleted files.
What is required it intelligence at the host level to provide the TP system with a list of blocks that are no-null but which the filesystem is no longer using.
This capability exists today in the form of the Thin reclamation API developed by Symantec.