I have run into problems running dbench with 80 clients
on a jfs partition.
Using 2.5.41-bk2 over the weekend, I ran dbench with up
to 128 clients on a single PIII, IDE machine running X.
Dbench runs on ext3, reiserfs, and xfs partitions were
completely successful. I was able to run dbench with
1,2,3,4,6,10,12,16,20,24,28,32,36,40,44,48,52,56,64,80,96,112
and 128 clients.
However on the jfs partition, the maxiumum dbench run was 64 clients.
At 80 clients, the dbench output stopped for many hours. This was
repeatable. Twice, the increasing client dbench script stopped at 80.
Here is a snippet from running ps with these options:
ps -ewo user,pid,priority,%cpu,stat,command,wchan
root 11058 15 0.0 DW [pdflush] lmGroupCommit
steven 11060 16 0.0 S time -v ./dbench wait4
steven 11061 15 0.0 S ./dbench 80 wait4
steven 11062 15 0.0 D ./dbench 80 TXN_SLEEP_DROP_LOCK
steven 11065 15 0.0 D ./dbench 80 TXN_SLEEP_DROP_LOCK
steven 11068 15 0.0 D ./dbench 80 lmGroupCommit
This was with dbench running on a jfs partition (/dev/hda11).
Filesystem Type Size Used Avail Use% Mounted on
/dev/hda1 ext3 236M 56M 168M 25% /
/dev/hda9 ext3 20G 7.3G 13G 38% /home
/dev/hda11 jfs 3.9G 478M 3.5G 13% /share_jfs
/dev/hda10
reiserfs 4.0G 37M 3.9G 1% /share_reiser
/dev/hda12 xfs 4.8G 253M 4.6G 6% /share_xfs
/dev/hda8 ext3 236M 4.7M 219M 3% /tmp
/dev/hda6 ext3 2.9G 1.3G 1.5G 47% /usr
/dev/hda7 ext3 479M 60M 395M 14% /var
I now find that if I try to ls the dbench directory on
/share_jfs, the ls command hangs. I was able to ls directories
above /share_jfs/steven/dbench earlier but now any ls command
on /share_jfs hangs. If I know the filename, I can cat the file,
for example my log of the dbench run, so reading of individual files
is still working.
Steven
On Tuesday 15 October 2002 10:29, Steven Cole wrote:
> I have run into problems running dbench with 80 clients
> on a jfs partition.
Steven,
I haven't run dbench with that many clients myself (or if I had, it's
been a while). I'm currently working on fixing some scalability issues
in JFS. In doing so, I'll try to reproduce your problem and make sure
it gets fixed.
Thanks,
Shaggy
--
David Kleikamp
IBM Linux Technology Center
On Tue, 2002-10-15 at 12:17, Dave Kleikamp wrote:
> On Tuesday 15 October 2002 10:29, Steven Cole wrote:
> > I have run into problems running dbench with 80 clients
> > on a jfs partition.
>
> Steven,
> I haven't run dbench with that many clients myself (or if I had, it's
> been a while). I'm currently working on fixing some scalability issues
> in JFS. In doing so, I'll try to reproduce your problem and make sure
> it gets fixed.
>
> Thanks,
> Shaggy
When I rebooted the test box to run 2.5.42-bk2, I said Y to running
fsck, and when fsck.jfs was run on that jfs partition, I saw one message
which indicated that a problem had been found. The message scrolled
away before I could save it. I immediately ran dbench with 64 clients
and then dbench with 80 clients on the jfs partition, this time
successfully. I'm now repeating the full ramp up dbench run, up to 128
clients. If nothing breaks, I'll shake it a little harder (like doing
several kernel compiles while dbench is trying to run).
Thanks,
Steven