Re: latch-free SCN scheme (10.1.0.3)

On Tue, 15 Feb 2005 14:42:34 -0800, yong321 wrote:

> I notice that prior to 10g, alert.log only says "SCN scheme 2" ("Picked
> latch-free" is not there). Not sure about its meaning. Probably a lazy
> way to say "Picked *mostly* latch free SCN ..." as the latch name
> suggests.
>
> In RAC, you still see that line immediately after startup. But when the
> database is opened, you see
>
> Tue Feb 15 14:08:37 2005
> ALTER DATABASE OPEN
> This instance was first to open
> Picked Lamport scheme to generate SCNs
>
> while the non-RAC database only says "alter database open".
>
> If I keep running
> select gets, dbms_flashback.get_system_change_number from v$latch where
> name = 'mostly latch-free SCN';
> I see that when SCN increments by 1, that latch get increments by 1; if
> SCN goes up by 2 or 3 or more, latch get also goes up but by a smaller
> number.
>
> I don't know what zero-copy redo is either but it looks like a 10g
> thing, and is controlled by _log_private_parallelism. There're also
> underscore parameters related to latch-free SCNs, even back to 8i.
>
> If you ever find an answer, please let us know.
>
> Yong Huang

Actually, I did come across the answer. There is an excellent article at: http://integrid.info/memory_latching.pdf The author is Tanel Poder, one of the first few Oracle Certified Masters.

-- 
Bug, noun: Son of a glitch