[Top] [All Lists]

Re: [PATCH 3/4] xfs: add quota-driven speculative preallocation throttli

To: Brian Foster <bfoster@xxxxxxxxxx>
Subject: Re: [PATCH 3/4] xfs: add quota-driven speculative preallocation throttling
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Mon, 17 Dec 2012 07:17:54 +1100
Cc: xfs@xxxxxxxxxxx
In-reply-to: <50CDDC74.6040302@xxxxxxxxxx>
References: <1354726078-31793-1-git-send-email-bfoster@xxxxxxxxxx> <1354726078-31793-4-git-send-email-bfoster@xxxxxxxxxx> <20121213022530.GF16353@dastard> <50CB500C.8090203@xxxxxxxxxx> <20121215212530.GD9806@dastard> <50CDDC74.6040302@xxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Sun, Dec 16, 2012 at 09:36:36AM -0500, Brian Foster wrote:
> On 12/15/2012 04:25 PM, Dave Chinner wrote:
> > On Fri, Dec 14, 2012 at 11:13:00AM -0500, Brian Foster wrote:
> >> On 12/12/2012 09:25 PM, Dave Chinner wrote:
> >>> On Wed, Dec 05, 2012 at 11:47:57AM -0500, Brian Foster wrote:
> >> Are you suggesting we
> >> set a default soft limit value on all quotas with a hard limit?
> > 
> > Yes. Your code already does this, effectively, by treating files
> > that have no softlimit set as always being over the soft limit.
> True, but only with respect to preallocation throttling. IOW, presumably
> there are other reasons to set (or not set) a soft limit from an
> administrative standpoint. Though I guess there's no reason to have a
> timer with no soft limit, so perhaps it doesn't matter. I'll play with it...

TO clarify, I'm talking about a soft limit only for preallocation
throttling purposes, not a general, implicit soft limit for

> To further clarify, is there any special behavior intended for a soft
> limit set in this manner? For example, the user creates quota, sets a
> hard limit and we set the new associated default soft limit. Should the
> new soft limit behave precisely as if the soft limit was set by the user
> from this point forward, or could it for example, remain volatile to the
> hard limit unless a soft limit value is explicitly set by the user
> (i.e., it is not written to disk and it is recalculated in memory if the
> filesystem is remounted; it is re-adjusted automatically if the hard
> limit is adjusted iff soft-limit hasn't been set explicitly, etc.).

See above - the implicit soft limit I suggested is only for prealloc
throttling.  Probaly better to call the two variabled
"q_prealloc_hard_limit" and "q_prealloc_soft_limit" or something
like that so that so it's explicit what their use is...


Dave Chinner

<Prev in Thread] Current Thread [Next in Thread>