Re: re-indexing per statement or per row?

And there was I thinking that you were making a very relevant observation that redo and undo can BOTH go take advantage of special behaviour to optimise performance by generating "block-level" records. Which is, in fact, the case when you have things like insert/select.

used_urec is indeed about undo/rollback records but you get interestingly similar anomalies in if you look at v$sysstat for "redo records".

I had got the significance of the "" marks around the 'know' - it is a convention I use myself to mean exactly the same thing. I was just absolving Howard from the blame because I know (without quotes) that there are several other sources where you might have got the information from.

--
Jonathan Lewis
http://www.jlcomp.demon.co.uk

Host to The Co-Operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/faq/ind_faq.html

Author of:
Practical Oracle 8i: Building Efficient Databases


Niall Litchfield wrote in message
<3cb2a04a$0$236$ed9e5944_at_reading.news.pipex.net>...


>"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message

>news:1018297158.13402.0.nnrp-14.9e984b29_at_news.demon.co.uk...

>> >This opens up a whole interesting can of worms for me because I *know*

>(cos

>> >the docs and folk like Howard have told me so) that *in backup mode*

that


>> >redo is generated for the whole block rather than the modified record(s)

>> >(assuming archivelog mode). Yet this is leading to a suggestion that

redo


>> >*can* be in someways a snapshot of block level changes. which raises the

>> >question of how ordinary redo differs from *backup* redo.

>> >

>>

>>

>> I doubt if Howard told you that, as it's not quite true.  In backup mode,

>> there is a very specific condition under which the entire current block

>> is copied into the redo buffer before the normal redo record is

generated.


>> And contrary to common belief, this extreme action is not taken for

>> every redo change generated for the block.

>

>My turn to clarify my statement - I had intended the I *know* above to be

>taken as an admission that what I knew was in fact provisional and probably

>wrong - it often is. What I understood from Howard is not necessarily what

>he has written :-( . What you say is relevant as I was a holder of the

>common belief that in backup mode whole blocks were always copied into redo

>it is interesting to learn that this is wrong.

>

>Having said all that the column in v$transaction that you point us at is

>used_urec or used undo records and undo is not redo as I know well, but

>managed to confuse above in any case. So unless used_urec actually referred

>to redo my whole argument above is completely baseless. Had to get that off

>my chest as I woke at 2 am this morning realising my foolishness.

>