Re: Oracle Myths- Tablespace placement answered by Oracle

"Nuno Souto" <nsouto_at_optushome.com.au.nospam> wrote in message news:3ce66661$0$15149$afc38c87_at_news.optusnet.com.au...
> In article <3ce51c7a$1_16_at_news.teranews.com>, you said (and I
quote):
> > > call me a dreamer, but if you put tables and indexes in same
> > tablespace
> > > and spread that tablespace over a number of devices you get
EXACTLY
> > the
> > > same result as above.
> >
> > You need to look down at the disk heads.
>
> Last thing in my list of priorities, Pablo! :-D
>
> > If we're issuing zillions of
> > single I/O's, that's a considerably different profile than
zillions of
> > I/O's that have some multi_block influence.
> > --
>
> It doesn't matter. At all. Think about it. An I/O is an I/O is an
I/O.

Actually, it does matter in that if the disk head is at one end of the platter and an index I/O sends it to the other end, that can be significant.

Now, Thomas points out the obvious (which I failed to do! <g>) that given b-trees, you're typically getting two levels of your b-tree cached. If you have a table with many many rows, you might see a b-tree with three levels and given random access to the table, you're going to have to issue one physical I/O to snarf the leaf.

--
Pablo Sanchez, High-Performance Database Engineering
mailto:pablo_at_hpdbe.com
http://www.hpdbe.com
Available for short-term and long-term contracts