Hi,
we are considering introducing quotas on our filer to prevent the "one user takes it all" situation, that happened several times and made the rest of the crew pretty unhappy.
Anyway, before doing that, I would like to check: Is there any "massive" performance degradation to be expected? Could someone out there share their real-life experience? Would be great.
I don't think that quotas impact performance very much. It is well worth having them.
Am I right that quotas should have an impact on write, only? And some extra CPU load?
Yes, quotas impact writes only, including things like removing files, and changing file ownership, of course. Besides CPU, quotas probably also eat up some RAM. I have no idea how much. During normal operation the overhead is worth the advantages of having quotas. Unless your CPU sits at near 100% all the time, I wouldn't worry about it.
Initializing quotas with the "quota on" command definitely chews up a lot of CPU, and it may run for minutes. But you should not have to do that very often. If you use default quotas then you should almost always be able to use "quota resize" (instead of "quota off" "quota on"). "quota on" looks at each file in the volume (just the inode, not the data) to initialize all the quotas, so the overhead is related to the number of files in the volume.
Generating a full quota report when you have a lot of users (we have approx. 25,000) also chews up a lot of CPU. You don't do this very often either. The overhead is related to the number of users, groups, and qtrees.
We use NFS, Clients are Solaris8. No qtree, just three appr. 400GByte Volumes?
Any hint to documentation and common pitfalls is highly welcome.
I would carefully study the na_quotas man page that specifies the format of the /etc/quotas file.
Bye, thanks wilfried
-- # Wilfried.Gaensheimer@Bingo-ev.DE, +49 841 975181 # # Room to rent ........................................#
Steve Losen scl@virginia.edu phone: 434-924-0640
University of Virginia ITC Unix Support