Re: commit times

  • From: Cary Millsap <cary.millsap@xxxxxxxxxxxx>
  • To: niall.litchfield@xxxxxxxxx
  • Date: Thu, 11 Feb 2010 22:33:43 -0600

Niall,

This is not exactly the format in which you've asked for the information,
but I think you'll find this interesting. Here is the output of
mrskew<http://method-r.com/component/content/article/117>on a
directory that contains a few dozen trace files from different
systems:

22:27:04 $ mrskew --name='log file sync' *.trc
  RANGE {min ≤ e < max}                  DURATION       CALLS
MEAN          MIN          MAX
-----------------------  ------------------------  ----------  -----------
-----------  -----------
   0.000000    0.000001           0.000000   0.0%
0
   0.000001    0.000010           0.000144   0.0%          27
0.000005     0.000004     0.000009
   0.000010    0.000100           0.004968   0.6%          82
0.000061     0.000043     0.000097
   0.000100    0.001000           0.147854  19.0%         348
0.000425     0.000100     0.000979
   0.001000    0.010000           0.481335  61.7%         150
0.003209     0.001015     0.008215
   0.010000    0.100000           0.145230  18.6%           5
0.029046     0.010715     0.095668
   0.100000    1.000000           0.000000   0.0%
0
   1.000000   10.000000           0.000000   0.0%
0
  10.000000  100.000000           0.000000   0.0%
0
 100.000000 1000.000000           0.000000   0.0%
0
1000.000000    Infinity           0.000000   0.0%
0
-----------------------  ------------------------  ----------  -----------
-----------  -----------
              TOTAL (5)           0.779531 100.0%         612
0.001274     0.000004     0.095668

22:27:10 $ mrskew --name='log file parallel write' *.trc
mrskew: no call names match /log file parallel write/i


Cary Millsap
Method R Corporation
http://method-r.com
http://carymillsap.blogspot.com


On Thu, Feb 11, 2010 at 5:20 PM, Niall Litchfield <
niall.litchfield@xxxxxxxxx> wrote:

> all
>
> on another list we've being discussing commit times. During this discussion
> the following assertion was made by some who knows full well what they are
> about and yet which surprised me. The quote is in relation to a reasonably
> heavily used system that incurs log file sync waits of between 15 and 20s on
> average.
>
> "At 17ms it had BETTER be an I/O problem. That sucks by the way - there's
> something horribly wrong if log file parallel write time is consistently
> >3ms "
>
> I've created a poll at  http://www.misterpoll.com/polls/473207  for both
> log file sync and log file parallel write, I'm intrigued as to what typical
> figures actually are. I'd be grateful if you could take a moment to look at
> a statspack or equivalent report from your monitoring tool of choice and
> update the survey with your results.
>
> cheers
>
> --
> Niall Litchfield
> Oracle DBA
> http://www.orawin.info
>

Other related posts:

  • » commit times- Niall Litchfield
  • » Re: commit times - Cary Millsap