Re: Increase in 'log file sync' waits while 'log file parallel write' remains unchanged
From: joel garry <joel-garry_at_home.com>
Date: Fri, 2 Oct 2009 14:38:17 -0700 (PDT)
Message-ID: <ce609289-862f-486c-88d1-85d92208eeb8_at_w37g2000prg.googlegroups.com>
On Oct 2, 12:49 pm, Srivenu Kadiyala <sriv..._at_hotmail.com> wrote:
> There is a bug in 9i due to which lgwr reports erroneous values for
> "log file parallel write".
> See Jonathan's note on this.http://www.jlcomp.demon.co.uk/log_file_writes.html
>
> Use OS tracing commands like truss to get the lgwr io timings.
>
> regards
> srivenu
Date: Fri, 2 Oct 2009 14:38:17 -0700 (PDT)
Message-ID: <ce609289-862f-486c-88d1-85d92208eeb8_at_w37g2000prg.googlegroups.com>
On Oct 2, 12:49 pm, Srivenu Kadiyala <sriv..._at_hotmail.com> wrote:
> There is a bug in 9i due to which lgwr reports erroneous values for
> "log file parallel write".
> See Jonathan's note on this.http://www.jlcomp.demon.co.uk/log_file_writes.html
>
> Use OS tracing commands like truss to get the lgwr io timings.
>
> regards
> srivenu
Ah, that must be why it sounds so familiar. Thanks!
But Jonathan's note does say "If you’ve ever been in the habit of discounting the log writer as a source of performance problems because the log writer times are good – think again. Any timings from the log writer are not to be trusted in 9.2. If you see high average wait times on log file sync, this may be the only way you can infer that you might have an I/O problem in the log writer." The increasing average wait times on log file sync is what the OP was asking about. I think we're all agreed that looking at what the OS is doing with I/O is what needs to be done.
jg
-- _at_home.com is bogus. There must be a Oracle Top Gun joke in here somewhere: http://www3.signonsandiego.com/stories/2009/oct/02/airborne-artistry/Received on Fri Oct 02 2009 - 16:38:17 CDT