Re: Strange load spikes on 2.4.19 kernel

From: Rob Mueller (robm@fastmail.fm)
Date: Sat Oct 12 2002 - 01:54:14 EST


> 1) If you don't need to know when a file was last accessed, mount the
> ext3 file system with the -noatime option. This disables updating of
> the "Last Accessed On:" property, which should significantly increase
> throughput.

Yes, already did that. Should have noted our fstab entry:

/dev/sda2 / ext3 defaults,noatime 1 1

> /sbin/elvtune -r 16384 -w 8192 /dev/mount-point
> where mount-point is the partition (e.g. /dev/hda5)

Thanks, we'll give this a try. Didn't know about this one before.

> 3) If the above don't work, double the journal size.

As you noted, we seem to be doing more reads than writes, so I'd be suprised
if 192M wasn't enough...

> P.S. You'd probably get more help from the ext3 mailing list.

I wasn't too sure that it was an I/O problem, which is why I posted here. As
the vmstat output showed, there didn't seem to be any sudden excessive I/O
occuring, but the load would jump enormously.

Maybe we should definitely try some different journaling modes, or disabling
journalling all together, to test if that is the actual culprit...

Rob

PS. I forgot to include the uptime dump in my last post. I just wanted to
show how spikey it was. Basically there's a long falling off decay, and then
a sudden spike again... which decays off... and then spikes again...
basically repeat what you see below over and over and over in 5-10 minute
intervals...

  1:29am up 23:50, 3 users, load average: 0.35, 2.49, 2.73
  1:29am up 23:51, 2 users, load average: 0.45, 2.44, 2.71
  1:29am up 23:51, 2 users, load average: 0.70, 2.42, 2.71
  1:29am up 23:51, 2 users, load average: 0.59, 2.34, 2.68
  1:29am up 23:51, 2 users, load average: 0.58, 2.28, 2.65
  1:29am up 23:51, 2 users, load average: 0.49, 2.21, 2.62
  1:30am up 23:51, 2 users, load average: 0.49, 2.15, 2.60
  1:30am up 23:52, 2 users, load average: 21.39, 6.43, 3.98
  1:30am up 23:52, 2 users, load average: 18.10, 6.22, 3.94
  1:30am up 23:52, 2 users, load average: 15.32, 6.01, 3.89
  1:30am up 23:52, 2 users, load average: 13.04, 5.83, 3.86
  1:30am up 23:52, 2 users, load average: 11.03, 5.64, 3.81
  1:31am up 23:52, 2 users, load average: 9.41, 5.47, 3.78
  1:31am up 23:53, 2 users, load average: 7.96, 5.29, 3.74
  1:31am up 23:53, 2 users, load average: 6.81, 5.13, 3.70
  1:31am up 23:53, 2 users, load average: 5.76, 4.96, 3.66
  1:31am up 23:53, 2 users, load average: 4.88, 4.80, 3.62
  1:31am up 23:53, 2 users, load average: 4.13, 4.64, 3.58
  1:32am up 23:53, 2 users, load average: 3.49, 4.48, 3.54
  1:32am up 23:54, 2 users, load average: 2.95, 4.34, 3.51
  1:32am up 23:54, 2 users, load average: 2.50, 4.19, 3.47
  1:32am up 23:54, 2 users, load average: 2.12, 4.05, 3.43
  1:32am up 23:54, 2 users, load average: 1.79, 3.92, 3.39
  1:32am up 23:54, 2 users, load average: 1.51, 3.79, 3.36
  1:33am up 23:54, 2 users, load average: 1.43, 3.70, 3.33
  1:33am up 23:55, 2 users, load average: 1.21, 3.58, 3.30
  1:33am up 23:55, 2 users, load average: 1.03, 3.46, 3.26
  1:33am up 23:55, 2 users, load average: 1.03, 3.38, 3.23
  1:33am up 23:55, 2 users, load average: 0.87, 3.27, 3.20
  1:33am up 23:55, 2 users, load average: 0.82, 3.17, 3.17

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Tue Oct 15 2002 - 22:00:42 EST