On Wed, Mar 7, 2012 at 1:46 PM, John R Pierce pierce@hogranch.com wrote:
On 03/07/12 11:27 AM, Warren Young wrote:
Why do those embedded systems have to see a whole 74 TB storage midden? Why not slice off a 1 TB or so LVM slice for them?
LVM management blows chunks. each of these arrays (we have dozens of these 74TiB storage servers) will end up nearly filled up. trying to manage 74 separate 1TiB file systems or even 10 separate 7TiB systems, and figuring out where there's freespace? no thanks. there's 1000s of those clients writing files to this archival storage, and the goal is to fill up the storage.
the operations folks at these production sites were using Solaris w/ ZFS for this in the previous (smaller) incarnation.
one solution proposed by a field site is to use SMB instead of NFS, as it doesn't seem to have these issues. I personally find SMB for (l)Unix to (l)Unix distasteful.
I can see where zfs would be good for this... Does this data have some sort of life cycle where you could give the devices a small space to write with the data periodically processed (server-side) into the large archive? Or could they use an http post, database write, or some other application-level protocol instead of seeing the filesystem directly?