From: Jonathan Lewis 05-Feb-03 08:33
Subject: Re : LMT Autoallocate Algorithm
I doubt if an algorithm will ever be published.
Every time someone asks the question the official
reply is that it is internal. This does mean that
it can change to address the possible honey-comb
effect you describe.
One minor detail - your experiments on sizes and rates
look as if they were based largely on a nice clean
tablespace. Oddities can happen (if in a clean start).
For example, try creating a table with (initial 64M) -
Oracle will create it with no 64K extents, and 64 x 1M
extents. Do the same with (initial 65M) and 8.1.7.4
and 9.2.0.2 will both create the first extent at 8M.
the next 40 something at 1M, and a couple more at 8M.
This makes it rather hard to guarantee that moving
a table, or exp/imp on a table will manage to fill
any funny little holes - possibly everything you
move will start allocating at larger sizes.
On the plus side - extents of size X are allocated on
a boundary matching X - in other words, they always
start in 'the right place' for their size. This does
reduce the probability of excessive honey-combing if
you have an early period of dropping and recreating
objects in the tablespace before things stabilise.
Jonathan Lewis
Author: Practical Oracle 8i