What platform is this? We had similar issue (PG 7.4.7). Raising number of checkpoint segments to 125, seperating the WAL to a different LUN helped, but it's still not completely gone. As far as disk I/O is concerned for flushing the buffers out, I am not ruling out the combination of Dell PERC4 RAID card, and the RH AS 3.0 Update3 being a problem. Thanks, Anjan
-----Original Message----- From: Donald Courtney [mailto:[EMAIL PROTECTED] Sent: Thu 5/19/2005 12:54 PM To: Tom Lane Cc: pgsql-performance@postgresql.org Subject: Re: [PERFORM] PostgreSQL strugling during high load Tom Thanks for the post - I think I am getting this problem for a synthetic workload at high connection loads. The whole system seems to stop. Can you give some examples on what to try out in the .conf file? I tried bgwriter_all_percent = 30, 10, and 3 Which I understand to mean 30%, 10% and 3% of the dirty pages should be written out *between* checkpoints. I didn't see any change in effect. /regards Don C. Tom Lane wrote: >"Mindaugas Riauba" <[EMAIL PROTECTED]> writes: > > >> It looks like that not only vacuum causes our problems. vacuum_cost >>seems to lower vacuum impact but we are still noticing slow queries "storm". >>We are logging queries that takes >2000ms to process. >> And there is quiet periods and then suddenly 30+ slow queries appears in >>log within the same second. What else could cause such behaviour? >> >> > >Checkpoints? You should ensure that the checkpoint settings are such >that checkpoints don't happen too often (certainly not oftener than >every five minutes or so), and make sure the bgwriter is configured >to dribble out dirty pages at a reasonable rate, so that the next >checkpoint doesn't have a whole load of stuff to write. > > regards, tom lane > >---------------------------(end of broadcast)--------------------------- >TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED] > > ---------------------------(end of broadcast)--------------------------- TIP 7: don't forget to increase your free space map settings ---------------------------(end of broadcast)--------------------------- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly