It shouldn't matter because all the extents are the same size.
You will lose some space if you create objects in parallel
because (on average) each PX slave will waste half an
extent as it finishes it's bit of the job. So, for example,
if you create an object using parallel 16 you will have
16 half empty extents in the object.

If the tablespace is freelist managed then the empty
space will be below the highwater mark and will be
scanned. (which means, I guess, that it may have to
be formatted and written at some point - I'll have to
check if Oracle has enhanced this part of the process).
If you use ASSM (bitmap freespace management) the
empty space will be skipped, but regions between the
"low highwatermark" and the "high highwatermark" may
be scanned in 16 block chunks.

You may want to create one very large object in two or three
ways and see how it behaves, just in case there is an obvious
advantage in one technology over the other.

My default choice would be locally managed, uniform at 8 - 64M,
with freelist management - unless I had some time to test before
going live. (And in your case you can always change your mind
within a 10 day cycle after you've gone live anyway).


Jonathan Lewis

Author: Cost Based Oracle: Fundamentals

The Co-operative Oracle Users' FAQ

Original Message -----
Cc: "Jonathan Lewis"
Sent: Wednesday, September 26, 2007 3:58 PM
Subject: Re:ASSM and tablespace fragmentation
thanks jonathan. If I am adding and dropping this much data do I have any risk
of fragmentation with uniform extents? Or does it not matter because all the
extents are the same size.
-------------- Original message ----------------------

Search Discussions

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
grouporacle-l @
postedSep 26, '07 at 4:35p
activeSep 26, '07 at 4:35p

1 user in discussion

Jonathan Lewis: 1 post



site design / logo © 2022 Grokbase