From: Andrew Morton <akpm@osdl.org>
Subject: Re: Updated ext4/jbd2 patches based on 2.6.19-rc1
Date: Fri, 6 Oct 2006 06:57:35 -0700
Message-ID: <20061006065735.1b51cc18.akpm@osdl.org>
References: <1160072610.8508.12.camel@kleikamp.austin.ibm.com>
	<20061005205526.7fe744f5.akpm@osdl.org>
	<20061005215442.310b7792.rdunlap@xenotime.net>
	<20061006055305.GG22010@schatzie.adilger.int>
	<20061005230401.0159e31b.akpm@osdl.org>
	<20061006064103.GJ22010@schatzie.adilger.int>
	<20061005235017.cbc4fdab.akpm@osdl.org>
	<m3k63divkw.fsf@bzzz.home.net>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Cc: Andreas Dilger <adilger@clusterfs.com>,
	Randy Dunlap <rdunlap@xenotime.net>,
	Dave Kleikamp <shaggy@austin.ibm.com>,
	ext4 development <linux-ext4@vger.kernel.org>
To: Alex Tomas <alex@clusterfs.com>
In-Reply-To: <m3k63divkw.fsf@bzzz.home.net>
Sender: linux-ext4-owner@vger.kernel.org

On Fri, 06 Oct 2006 14:31:59 +0400
Alex Tomas <alex@clusterfs.com> wrote:

> >>>>> Andrew Morton (AM) writes:
> 
>  AM> On Fri, 6 Oct 2006 00:41:03 -0600
>  AM> Andreas Dilger <adilger@clusterfs.com> wrote:
> 
>  >> The big performance win will come with mballoc and delalloc.  CFS has
>  >> been using mballoc for a few years already with Lustre, and IBM + Bull
>  >> did a lot of benchmarking on it.  The reason it isn't in the first set of
>  >> patches is partly a manageability issue, and partly because it doesn't
>  >> directly affect the on-disk format (outside of much better allocation)
>  >> so it isn't critical to get into the first round of changes.  I believe
>  >> Alex is working on a new set of patches right now.
> 
>  AM> Are you sure that these things will improve allocation much?  Reservations
>  AM> made a big improvement there.
> 
> it depends on underlaying storage and workload. mballoc uses buddy
> internally. it's much simpler and cheaper to find free 2^N blocks
> compared to bitmap.

So mballoc's application is to save CPU cycles?

> this is especially important for arrays like
> DDN and raid5/6 because they require stripe-aligned/-sized requests
> for good throughput.

Does this not imply that there needs to be new linkage between the
filesystem and the lower layers?  So that raid/etc can inform the
filesystem driver about its alignment and striping requirements?

> also, last mballoc takes logical block into
> account and can preallocate few chunks at different logical offsets
> for a file. imagine torrent downloading different pieces from few peers.

hm.  You don't need anything as exotic as bittorrent to show up problems in
that area:

box:/usr/src/25> sudo bmap vmlinux | wc -l
1152