Re: Myth revisited ...

"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message news:<bpaocc$ov9$1$8302bc10_at_news.demon.co.uk>...
> I don't think multiple block sizes really changes
> the argument. I think the line that was generally
> accepted after the last round of discussion was
> that you separated objects because of different
> characteristic behaviour.
>
> If it is the characteristic of an object that it's
> behaviour would benefit from being in a tablespace
> with a different block size, then it goes in a
> different tablespace. It's just another dimension
> to categorising objects and their behaviour
> patterns.

I can't help but think that a major characteristic of indices is schizoid behavior. How do you categorize something that sometimes is typically online, gimme-that-particular block then other times is range scanned? While that is true of tables too, don't indices push this enough further (on density, if nothing else) to be qualitatively different?

It's tempting to think indices should have their own ts for performance, _because_ they are different in this way. Being able to identify index-specific hot-spots would be a bonus.

>
> I would object very strongly to any ideas like:
> Separate tables from indexes because
> tables should be in 8K blocks and indexes
> should be in 16K blocks.
> which is the sort of area that people would be
> heading towards if they got carried away with
> the question you raised.
>
>

Thanks especially for the other post pointing out why not to use 32K index blocks and that it might detrimentally affect CBO sort/merge decisions.

jg

--
@home.com is bogus.
http://www.davemcnally.com/lyrics/KingCrimson/21stCENTURYSCHIZOIDMANincludingMirrors.asp