On Sat, 24 Nov 2007, Guillaume Smet wrote:
Here are some rough results: http://people.openwide.fr/~gsmet/postgresql/postgresql_8.3_development_cycle_1.png
So the big dips were Jan->Feb and April->May. I've still got the text of the commit log Tom assembled sitting at http://developer.postgresql.org/index.php/8.3_Changelog and http://www.westnet.com/~gsmith/content/postgresql/83revhist.txt if anyone wants a relatively easy way to see what happened then.
Scanning the January results, the biggest drop, I noted that 2007-01-16 was the day autovacuum was turned on by default. If you're sure it's explicitly disabled in every test that shouldn't matter, might be something worth double-checking.
The other commits that looked like a lot of code changes with potential for performance changes were all from Tom, on 2007-01-03, 2007-01-08, and 2007-01-10, and 2007-01-20. If you wanted to try and nail this down further, three dates with no commits that would narrow the source of the big regression(s) that month are 2007-01-07, 2007-01-13, 2007-01-22. 2007-01-17 would be a good additional day to confirm that autovacuum wasn't responsible.
packed varlenas were commited during April
There were also several changes to pgbench itself that month. Useful breakpoints in that month to subdivide might be 2007-04-08 (after varlena change), 2007-04-17, 2007-04-22.
-- * Greg Smith [EMAIL PROTECTED] http://www.gregsmith.com Baltimore, MD ---------------------------(end of broadcast)--------------------------- TIP 6: explain analyze is your friend
