Date: Wed, 4 Apr 2001 19:03:46 -0400
From: Michael MacDonald <mjmac@ximian.com>
To: linux-xfs@oss.sgi.com
Subject: reproducible wedge/corruption
Message-ID: <20010404190346.D6693@dr-zaius.ximian.com>
Reply-To: mjmac@ximian.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.3.12i
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk

Hi all.  I've been lurking for a while, trying to get a feel for how XFS
is working for people.  It seems to be working well overall, which is a
little disappointing for me because I'm not having much luck with it.  
Perhaps I'm not living right, or something.

Anyhow, I have two machines with similar hardware on which I can pretty
reliably cause the system to lock solid.  I've tried different kernels
(linux-xfs-beta vs. linux-xfs; 2.4.1 -> 2.4.3), but the problem doesn't 
go away.  The kernel is configured with the bare minimum to run an 
IDE-based server.

To try to reproduce the error that users were running into, I wrote a
simple script to pound on the fs.  It hasn't made it past the first or
second iteration on an XFS partition, because the machine will hang.
Can't log in at the console or anything.  I tried changing the logbsize
parameter as described in the FAQ, and got better performance until the
machine hung like it did before.  I gave up after several days of
tweaking and reformatted ext2.  I would wonder if I have something
misconfigured except for the fact that the script runs through all 30
iterations on an ext2 filesystem without a hang.

I'll post the script if requested, but basically all it does is:
create 10000 small files in one directory
make 10000 symlinks in another directory to the files in the first
remove the second directory
remove the first directory

It seems to be hanging mostly during the removal stage, although I have
seen it hang during file creation.  On two of the hangs, I have seen
some nasty file corruption.

The specifics of the system are as follows:
Dell PowerEdge 350
Celeron 600
128MB RAM
Intel 440BX chipset

I've got another 350 with
PIII 700
512MB RAM

They both seem to exhibit similar behavior wrt XFS, although I haven't
pounded on the second one quite as much.

If anyone has any ideas, or would like more details, please let me
know.  I would really like to get XFS on these things, as they've got a
lot of disk between the two of them.
-- 
Michael MacDonald       Systems Monkey
mjmac@ximian.com        Ximian, Inc.