Re: 9i multi cache buffer

"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message news:ar5po0$irt$1$8302bc10_at_news.demon.co.uk...
>
> The old biblical one about 'motes and beams' (King James
> version - possibly 'sawdust and planks' in more modern
> versions) springs to mind.
>
> Whilst Oracle publishes tpc's that appear to get
> performance benefit from multiple block sizes I
> think most users will take the (wrong) hint and
> depend on the feature for performance benefits -
> whilst failing to notice other, far more significant,
> options for performance improvement.
>
>
> Here's a thought for performance, though.
> If you have a 2K block size for UNDO on
> an OLTP system - every undo block is likely
> to waste about 400 bytes i.e. 20%. If you
> change to 16K blocks, that's only 2.5%,
> which means more efficient use of buffer
> space - hence less I/O, hence ... Of course,
> I can come up with three reasons why this
> apparent performance enhancer will lead to
> worse performance - dependent on the exact
> nature of your application and installation !
>
> Remember the Xerox salesman's cry:
> "Sell benefits, not features"
>
>
> --
> Regards
>
> Jonathan Lewis
> http://www.jlcomp.demon.co.uk
>
> Next Seminar dates:
> (see http://www.jlcomp.demon.co.uk/seminar.html )
>
> ____USA__________November 19/21 (Dallas)
> ____England______January 21/23
>
>
> The Co-operative Oracle Users' FAQ
> http://www.jlcomp.demon.co.uk/faq/ind_faq.html
>
>
>
>
>
> Mark Townsend wrote in message ...
> >
> >The reality is that for most workloads (and I'd dare to say all
> customer
> >workloads), the performance benefits from multiple block sizes will
> not make
> >a significant impact, and in fact, I doubt would be measurable at
> all.
> >
> >Hopefully this will close this thread (as it did the one that went
> through
> >Oracle about 6 months ago on this same topic)
> >
>
>
>

All of which is why, when people ask "What block size should I use?", I answer:

1. Insufficient data. What are your objectives/OS platform/DB version/application characteristics?

or

b) It depends

or (failing everything else)

8K.

Over-simplification, I know, but it answers the question.

In any case, in real life it's usually the crappy SQL that makes the difference.

Regards,
Paul Received on Sun Nov 17 2002 - 13:59:49 CST