What we traditionally do is apply a quota, but make it absurdly high — back in the day of 500GB filers, everyone had a 32GB quota. Why? Because quotas are updated in near-time, so you can see who is filling up disk space by just running a repquota -a (or equivalent) and you get a fast summary of who’s using what. Wait five minutes, then do it again, and compare the results.

This falls down in a development environment, where people can write to their home directories as other people — one company I have lets root on development nodes write to home directories, which means that “nobody” (or “nobody4″) owns a lot of files that we can’t see that way, so your milage may vary…

(I actually presented this at a tech conference once – got the slides somewhere).

Never allocate all your disk. Keep a fraction (10-20%) unallocated. When disk grows unexpectedly & creates a crisis, do the following:

(1) Go to your boss with an emergency request for $10k to add on to the disk pool. A blank PO is best, you never know where you'll have to go to find lots of disk on short notice.

(2) After you get the money, extend your disk into the unallocated space and resolve the crisis.

(3) Take the money & buy a big monitor and a cool new laptop.

Of course I'm the boss now so that won't work in my shop.

–Mike

That’s a cool concept. Locking down the partition before the users can thrash it.

I’ve got nagios alerts triggering at 90% (typically) with a warning, but it would be nice to not worry about the emergency run-n-dump when the partition fills up

Each directory at the root level had a hard quota smaller than the volume, with some overlap in cases where there was more than one directory at the root of the volume. That quota existed only to prevent userland from filling the volume and making a mess out of the volume.

Then we still had normal per users/per volume quotas, but we were fairly liberal about the quotas.

–Mike

# ls -d Accounts/* | wc -l
27

# du -hs Accounts/
46G Accounts/

That’s just email.

/sigh

However, we do limit mail box sizes. Alloted size depends on job title, IE: CE_, VP, RVP, Director, Manger, “everyone else” and ranges from 150mb to 500mb, IIRC. All mail over 90 days old is deleted from the server, allthough it does get stored off the server for a longer period of time, depending on government regulations and such, in the case of a legal situation.

That’s interesting. I’d be nervous about that in this instance, just because the entire partition isn’t dedicated to the user directories. I think that might be the sticking point. Ideally I should make the home directories their own partition, but I’ve got them on a /mnt/data that also has some general storage stuff on it.

I’ll have to reconsider how I’m doing it, I suspect.

Do you ever have issues with users blitzing you with huge runs on diskspace in short amounts of time?

Depending on how many users you have (and how much wiggle room you have on your disks), you might find hard quotas to be a bit of a pain. Nearly half of the Linux support tickets I get are fixed by reducing quota usage — especially login problems and Open Office errors.