Re: lots of interesting stuff from everybody about quotas.

On our toasters we have a huge mess called 'home directories' and another huge mess called 'project areas'. We've glass-walled these various areas with qtrees, thus making ersatz filesystems that behave to clients exactly like filesystems except to the filer and admin host.

However, many users 'live' and 'work' in the same areas as each other, so a qtree quota on the filer wouldn't necessarily map to any one user's quota. In fact quite the reverse.

Quotas, in the Unixy sense are distinct from this, because they would report information about each numeric uid's usage of space in a given filesystem and so would have to be specified in a table for each qtree, mapping:

uid->(soft, hard, timelimit, timeleft {, any other useful stuff})

It would be a nice interface-feature to be able to harness qtree quotas to filer-wide quotas, or even groups-of-filers-wide quotas, but only on demand (!) as the control of each qtree quota table would be more generally useful. Perhaps users could be given default quotas for all qtrees on a filer until they appear in specific tables which override the defaults.

The provision of soft and hard quotas per qtree for defined sets of uid's would be a real admin bonus for us.

BTW I was under the impression that a machine hosting a quota'd filesystem locally and serving NFS, was able to deny space to a client that didn't observe quotas if a transaction would breach the quota conditions..? I think I need to go find out some more, as this sounds reasonable but doesn't gel with what others have prior to this on this discussion.

How quotas per user under NFS would map to CIFS quota'ing (does such a beast exist?) I've no idea yet, though I can see a clear need to have the two work synergistically. (Users who flit from Unix to PC and back using all the same files at the same time make life interesting that way).

(dang, another 2p gone)