xfs
[Top] [All Lists]

Re: [PATCH 0/18] xfs: metadata and buffer cache scalability improvements

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 0/18] xfs: metadata and buffer cache scalability improvements
From: Alex Elder <aelder@xxxxxxx>
Date: Tue, 21 Sep 2010 11:23:12 -0500
Cc: XFS Mailing List <xfs@xxxxxxxxxxx>
In-reply-to: <20100921020203.GC2614@dastard>
References: <1284461777-1496-1-git-send-email-david@xxxxxxxxxxxxx> <1284729700.5524.53.camel@doink> <20100921020203.GC2614@dastard>
Reply-to: aelder@xxxxxxx
On Tue, 2010-09-21 at 12:02 +1000, Dave Chinner wrote:
> On Fri, Sep 17, 2010 at 08:21:40AM -0500, Alex Elder wrote:
> > On Tue, 2010-09-14 at 20:55 +1000, Dave Chinner wrote:
> > > This patchset has grown quite a bit - it started out as a "convert
> > > the buffer cache to rbtrees" patch, and has gotten bigger as I
> > > peeled the onion from one bottleneck to another.
. . .

> 
> All together, with the LRU code being reworked a bit w.r.t. stale
> buffers and shrinker behaviour.
> 
> In reality, though, i don't think that separating them into separate
> series make much sense. The order they are in right now is
> bisectable and fairly logical....

I have been thinking about this since sending it.  I think my
concern was not so much that they were all in one series.  It's
more about the fact that you are doing a number of non-trivial
changes, all together.  And as such my perception of the combined
risk of committing them all at once is higher.  So what I was
probably after was somehow being able to verify each chunk of
the series separately, spilling them out gradually rather
than all at once.

But in the end, I guess I agree with what you say.  If we could
get some parts--like those you say are standalone--committed
earlier (and then out for wider exposure sooner) that would be
good, but otherwise it's OK as a single series.  I'll look for
your next update, and will just wait for pull request(s) when
you feel they're ready.

                                        -Alex

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