Re: latch-free SCN scheme (10.1.0.3)
- From: Martic Zoran <zoran_martic@xxxxxxxxx>
- To: ntufar@xxxxxxxxx, tanel.poder.003@xxxxxxx
- Date: Wed, 16 Feb 2005 01:12:43 -0800 (PST)
My way of thinking is also that there is no need to
copy anything to log buffer if the strands are created
in the unique static way. LGWR should know about them
and their location and who did commit. Maybe in some
new x$ table.
But then what are shared strands? Who is sharing them
in the single instance? Are they for distributed
transactions or something else?
Are these for all others that cannot aquire any
private one, like old log_buffer area?
Then maybe the new log buffer structure is just
consists of private and share strands, no copy, then
LGWR will know which strand(s) to write to the disk.
Gone too far ....
Regards,
Zoran
--- Nicolai Tufar <ntufar@xxxxxxxxx> wrote:
> On Wed, 16 Feb 2005 00:56:05 -0000, Tanel P=F5der
> <tanel.poder.003@xxxxxxx>=
> wrote:
> > Btw, this description of zero-copy redo is just my
> untested guess, based =
> on
> > few parameter names, latch statistics and
> background information. Even
> > though operating systems do use such VM page
> remapping tricks for (like s=
> ome
> > TCP stack implementations) for performance reasons
> - Oracle might not do
> > it - zero copy might also mean that redo records
> are directly constructed=
> to
> > private strand areas and written to disk directly
> from there by logwr, as
> > opposed to creating these in PGA and copying to
> logbuffer then...
>
> VM page remapping is not possible for a user
> process. I speak for Solaris and Linux. Other OSes
> may
> allow it though. Anyways, since Oracle is largely
> UNIX-
> oriented, your second scenario is more likely.
>
> > Tanel.
> Nicolai
> --
> //www.freelists.org/webpage/oracle-l
>
__________________________________
Do you Yahoo!?
Yahoo! Mail - Easier than ever with enhanced search. Learn more.
http://info.mail.yahoo.com/mail_250
--
//www.freelists.org/webpage/oracle-l
Other related posts:
