Zdenek Kotala wrote:
It looks likes that there are lot of lock issues on FSM pages. When number of FSM pages is increased then number of collisions is lower. It is probably why 2 clients significantly speed up between 33MB and 333MB.
Yes, that's what I thought as well. With table size under 33 MB, the FSM consists of just one (bottom-level) FSM page,
I think it is time to take DTrace ;-).Do you have any machine with DTrace support?
No.
If not send me your test suit and I will try it run on my machine.
Sure, here you are. tests.sh is the main script to run. You'll need to adjusts the paths there for your environment.
As it is, the tests will take many hours to run, so you'll probably want to modify tests.sh and pgbenchtests.sh to reduce the number of iterations. At least on my server, the variance in the numbers was very small, so repeating the tests 4 times in tests.sh is probably overkill.
Thanks! -- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com
fsmtest.tar.gz
Description: GNU Zip compressed data
-- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
