Re: different physical access method because of disabling Automated Memory Management?
- From: "Alex Fatkulin" <afatkulin@xxxxxxxxx>
- To: martin.a.berger@xxxxxxxxx
- Date: Sat, 3 Jan 2009 14:25:17 -0500
Martin,
I did some quick and dirty ad-hoc tests recently and thus far it seems
to certainly depend on:
1. _small_table_treshold as Tanel has mentioned
2. Amount of table blocks already presented in the buffer cache
3. Amount of dirty blocks (since we need to do an object level
checkpoint for this)
(2) seems to be 1/2 of a table blocks and (3) 1/4 of a table blocks,
you can see my test at http://afatkulin.blogspot.com
the main problem here is that amount of variables that Oracle could
use as in input to make a decision is probably much larger than the
above three I've mentioned thus it is what it is -- just documenting
some observations without really approaching any reliable figures.
On Sat, Jan 3, 2009 at 3:35 AM, Martin Berger <martin.a.berger@xxxxxxxxx> wrote:
> Hi Tanel,
> I checked _small_table_treshold (for some reasons it was not mentioned in
> the 10053 trace.)
> You where right, there was a difference:
> manual:
> =======
> _small_table_threshold => 59
> auto:
> ====
> _small_table_threshold => 498
> So I adjusted my 'manual'-testcase, run it again - but still 'direct path
> read'.
> Does anyone knows which subsystem takes the decision which kind of IO, and
> how to trace this?
> best regards,
> Martin
> --
> Martin Berger
> http://berxblog.blogspot.com
>
>
--
Alex Fatkulin,
http://afatkulin.blogspot.com
http://www.linkedin.com/in/alexfatkulin
--
//www.freelists.org/webpage/oracle-l
Other related posts:
