Date: Thu, 9 May 2002 08:56:07 +1000
From: Nathan Scott <nathans@sgi.com>
To: Eric Sandeen <sandeen@sgi.com>
Cc: Austin Gonyou <austin@coremetrics.com>, linux-xfs@oss.sgi.com
Subject: Re: TAKE - pagebuf
Message-ID: <20020509085607.E131535@wobbly.melbourne.sgi.com>
References: <200205080421.OAA52473@snort.melbourne.sgi.com> <1020871711.19190.18.camel@UberGeek> <1020874330.30053.2.camel@stout.americas.sgi.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.2.5i
In-Reply-To: <1020874330.30053.2.camel@stout.americas.sgi.com>; from sandeen@sgi.com on Wed, May 08, 2002 at 11:12:09AM -0500
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk

On Wed, May 08, 2002 at 11:12:09AM -0500, Eric Sandeen wrote:
> On Wed, 2002-05-08 at 10:28, Austin Gonyou wrote:
> > Curious to know what is being considered *small block*? (2k 1k?)
> 
> Anything less than page size.
> 
> And "boost performance" does not mean that reducing your block size is a
> performance enhancer; the mod you mentioned was more along the lines of
> "less slow," not "super fast."

This mod gets <pgsize blks up to about the same as pgsize for specific
workloads/machine configs (it is a huge boost over the previous code).
So, it's better than less slow, but not quite super fast yet. ;)  But,
its all academic at this stage, since...

> Note that none of this stuff is ready for prime-time yet, either... You
> should assume that the block size == page size restriction is still in
> place, for now.

It _is_ still in place (the Linux/XFS mount code path doesn't allow fs's
with non-pgsize blocks), no assumption about it.  There is still plenty
to be done before we drop this check I think, but not too far off now.

cheers.

-- 
Nathan