Re: Is there a performance limit of extents?

There is a little room for hand-waving in this area when using parallel query. (Though the 'cache an extent' line is incorrect).

When calculating the rowid ranges to pass to parallel query slaves, there is an algorithm that accumulates contiguous extents, then divides them up and passes them out on
a nominal 9:3:1 ratio.

If you have large numbers of relatively small extents which are NOT adjacent to each other then the Query co-ordinator can end up passing out a very large number of very small chunks to query slaves, which can result in a silly amount of overhead. It is therefore sensible to aim for a relatively low number of extents per segment on objects likely to be targeted by Parallel Queries.

--

Jonathan Lewis
Yet another Oracle-related web site:  http://www.jlcomp.demon.co.uk

Thomas J. Kyte wrote in message <8dkg4r$jl5$1_at_nnrp1.deja.com>...


>In article <956092535.16586.0.pluto.d4ee154e_at_news.demon.nl>,

>  "Sybrand Bakker" <postbus_at_sybrandb.demon.nl> wrote:

>

>> Also, Oracle typically tries to process an extent as one unit, cache

>> it completely in memory etc.

>

>no we don't.  we cache blocks and such.  extents (which may be upto

>2gig in size) will not be cached as an entity.

>