Thoughts on Quotas

Michael Neumann mneumann at ntecs.de
Wed Sep 29 11:34:24 PDT 2010


Am Mittwoch, den 29.09.2010, 19:54 +0200 schrieb Jasse Jansson:
> On 09/29/10 03:47 PM, Rumko wrote:
> > Stathis Kamperis wrote:
> >
> >> 2010/9/28 Rumko<rumcic at gmail.com>:
> >>> Stathis Kamperis wrote:
> >>>
> >>>> 2010/9/28 Sdävtaker<sdavtaker at gmail.com>:
> >>>>> What i tried to sai about history was that user usage should be
> >>>>> measured in a different bag than the history that the user usage
> >>>>> generated.
> >>>>> Sorry if it was not clear, english is not my main language and i use
> >>>>> to fail time to time. :-/
> >>>>> Damian
> >>>>>
> >>>> I kind of agree.
> >>>>
> >>>> Why "punish" user for something that s/he is not able to control
> >>>> directly ? Even more, the user may not be aware of the underlying
> >>>> filesystem's technicalities (how it retains history and so on).
> >>>>
> >>>> Better come up with something else.
> >>>>
> >>>> Best regards,
> >>>> Stathis
> >>> Not punishing that user means punishing the whole system and everything
> >>> depending on that system. And as I said before, it's user's data, so who
> >>> should be punished if not the user? The user can always tell the admin that
> >>> he does not any history at all or how much history he needs, so it's purely
> >>> that user's responsibility ... his data, his rules, his reponsibility.
> >>> [...]
> >> Ok, fine. I'm not strongly opinionated on this. I'm just thinking from
> >> the Josephine perspective, who may not (or even want to) know how her
> >> file-system operates.
> > Then that user's home dir can be nohistory and there is no problem?
> > If she doesn't need multiple copies of her data, then I see no reason why to
> > keep that data. But if she wants n snapshots/backups then there should be a
> > limit on how much total space she can take. Otherwise we could just display
> > du -h of her home dir when she logs in and it would be about as useful at
> > limiting her disk usage, so there would be no point to a quota ;)
> > On purpose she could bring down the whole system at will even though she
> > was "limited", but unfortunately she could do it unknowingly as well
> > (downloading flash videos, powerpoint jokes, maybe a movie or two, etc. over
> > and over and rewriting the old files).
> >
> >> But if we go this route, then we should also provide history retention
> >> statistics to user-land utilities, such as df(1).
> >>
> >> Imagine the confusion of a user that types 'df', sees that the quota
> >> threshold hasn't been reached, yet she is denied further disk storage
> >> allocation.
> > Agreed, now we just need to find someone to do it :P
> 
> While you're at it, why don't make two kinds of snapshots.
> 
> 1. A user initiated snapshot, usnap, that the user controls and counts 
> towards the quota limit.
> 
> 2. System snapshots, ssnap, obviously managed by the system/admin (out 
> of control of the user) and therefore counts as system overhead.

Well, that would be doable but would need some hacking in hammer
cleanup. And the cleanup/reblock process does not know about file names,
only inode numbers AFAIK, so that every usnap would end up in generating
a full PFS snapshot. It is clearly doable, but would probably be very
inefficient. But, hmm, telling "hammer cleanup" to throw away history
for "*.bak" or other "unused" files would be nice :). But then,
snapshots right now are beautiful as they are so easy to use and you
cannot make much wrong.

Regards,

  Michael






More information about the Kernel mailing list