Re: Temporary table use and the redo log
Date: Mon, 14 Sep 2009 04:29:31 -0700 (PDT)
On Sep 13, 1:15 pm, "Jonathan Lewis" <jonat..._at_jlcomp.demon.co.uk> wrote:
> Running on 22.214.171.124 (where the bug is fixed) the insert reports about
> 38 redo records and 32KB of redo size on a global temporary table
> (compared with about 65 and 100KB on a normal table). This is due
> to the way that Oracle can pack multiple changes to a single block
> into a single undo record and redo change vector).
> On 126.96.36.199, this test reported 2,000 redo records and about 350KB
> or redo generated because that packing optimisation was missing.
> According to my notes, this bug was fixed in 188.8.131.52 (so my comment
> about "most" versions of 9i was wrong) and 10.1.0.4. (The optimisation
> was in 8i).
thanks for the explanation. Your test case seems to refer to conventional DML, does this also apply to direct-path inserts? The test case would generate undo and redo for the primary key index, so I could imagine that it applies there, too, but I'm not sure.
I don't have a pre-184.108.40.206 version at hand (and no 10.1 either), so cannot test myself at present.
Oracle related stuff blog: