Re: flashback buf free by RVWR waits and LOB caching
From: Tanel Poder <tanel_at_poderc.com>
Date: Mon, 28 Feb 2011 15:16:25 +0200
Message-ID: <AANLkTimHi0rMdkV+7fPypDm6D-rsCEqw0VrjgSA5Cne8_at_mail.gmail.com>
On Mon, Feb 28, 2011 at 10:25 AM, <Laimutis.Nedzinskas_at_seb.lt> wrote:
Date: Mon, 28 Feb 2011 15:16:25 +0200
Message-ID: <AANLkTimHi0rMdkV+7fPypDm6D-rsCEqw0VrjgSA5Cne8_at_mail.gmail.com>
On Mon, Feb 28, 2011 at 10:25 AM, <Laimutis.Nedzinskas_at_seb.lt> wrote:
> ...
>
> Only that the actual recovery would happen when LOB is first accessed.
> Which brings some interesting "hows" how Oracle performs LOB version of
> "block cleanout" but again, that is doable provided enough data is stored
> for crash recovery. The only questions are where that data is stored and
> how it affects performance.
>
>
LOB read consistency & block versioning is done using LOB indexes. They keep
pointers to different versions of LOB chunks in them... I have a little
illustration on page 21 of my old LOB internals presentation:
http://www.slideshare.net/tanelp/oracle-lob-internals-and-performance-tuning
Tanel
-- http://www.freelists.org/webpage/oracle-lReceived on Mon Feb 28 2011 - 07:16:25 CST