Hi Michael,
Michael Fuhr wrote:
> If you run pgbench several times without intervening checkpoints,
> do your postmaster logs have any messages like "checkpoints are
> occurring too frequently"? It might be useful to increase
> checkpoint_warning up to the value of checkpoint_timeout and then
> see if you get any such messages during pgbench runs. If checkpoints
> are happening a lot more often than every checkpoint_timeout seconds
> then try increasing checkpoint_segments (assuming you have the disk
> space). After doing so, restart the database and run pgbench several
> times without intervening checkpoints and see if performance is
> more consistent.
I got the "checkpoints are occurring too frequently". Increasing the number of checkpoint_segments from the default 3
to10 resulted in more tests without performance penalty (~ 5-6 tests). The perfomance penalty is also a little less. It
takesseveral minutes for the background writer to catch up.
This will solve my problems at the customers site (they do not run sm many sales transaction per second), but not my
ownproblem while converting the old database to a new databse :-(. Maybe I should invest in other hardware or
re-arrangemy RAID5 in a RAID10 (or 50???).
Groeten,
Joost Kraaijeveld
Askesis B.V.
Molukkenstraat 14
6524NB Nijmegen
tel: 024-3888063 / 06-51855277
fax: 024-3608416
e-mail: J.Kraaijeveld@Askesis.nl
web: www.askesis.nl