Re: Excessive transaction tables consistent reads - undo records applied in 11.2.0.3
From: Tanel Poder <tanel_at_tanelpoder.com>
Date: Thu, 4 Oct 2012 19:54:58 -0700
Message-ID: <CAMHX9JLd7UONp3zVsMmjXiO3=cyx4LDRfCF9dkP=JvsWFBr5Vg_at_mail.gmail.com>
On Thu, Oct 4, 2012 at 8:37 AM, Administrator User < jonathan_at_jlcomp.demon.co.uk> wrote:
>
> Part of the problem, of course, is that Oracle starts a tablescan from the
> beginning while typical batch processing is all about changing the most
> recent data (i.e. the stuff at the end of the table). So the time you spend
> reading the first 350 MB may be letter the other session wreck the last
> 50MB. (This is a topic close to Mark Farnham's heart - why doesn't Oracle
> give us a "tablescan descending" hint.)#
>
>
If you don't mind occasionally getting wrong results (as this is undocumented, unused and unsupported feature), you can use event 10460 "Perform backward tablescans for consistent read reduction" for backwards full table scans ;-)
Date: Thu, 4 Oct 2012 19:54:58 -0700
Message-ID: <CAMHX9JLd7UONp3zVsMmjXiO3=cyx4LDRfCF9dkP=JvsWFBr5Vg_at_mail.gmail.com>
On Thu, Oct 4, 2012 at 8:37 AM, Administrator User < jonathan_at_jlcomp.demon.co.uk> wrote:
>
> Part of the problem, of course, is that Oracle starts a tablescan from the
> beginning while typical batch processing is all about changing the most
> recent data (i.e. the stuff at the end of the table). So the time you spend
> reading the first 350 MB may be letter the other session wreck the last
> 50MB. (This is a topic close to Mark Farnham's heart - why doesn't Oracle
> give us a "tablescan descending" hint.)#
>
>
If you don't mind occasionally getting wrong results (as this is undocumented, unused and unsupported feature), you can use event 10460 "Perform backward tablescans for consistent read reduction" for backwards full table scans ;-)
-- Tanel Poder Blog - http://blog.tanelpoder.com App - http://voic.ee -- http://www.freelists.org/webpage/oracle-lReceived on Fri Oct 05 2012 - 04:54:58 CEST