Message-ID: <437D13AA.4010309@openwide.fr>
Date: Fri, 18 Nov 2005 00:35:06 +0100
From: Guillaume Smet <guillaume.smet@openwide.fr>
User-Agent: Mozilla Thunderbird 1.0.6-1.1.fc3 (X11/20050720)
MIME-Version: 1.0
To: Andrew Sullivan <ajs@crankycanuck.ca>
CC: pgsql-performance@postgresql.org
Subject: Re: weird performances problem
References: <437CC21D.8070402@openwide.fr>
	<20051117221314.GC26696@phlogiston.dyndns.org>
In-Reply-To: <20051117221314.GC26696@phlogiston.dyndns.org>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Andrew,

Andrew Sullivan wrote:
 > Is it exactly half an hour?  What changes at the time that happens
 > (i.e. what else happens on the machine?).  Is this a time, for
 > example, when logrotate is killing your I/O with file moves?

No, it's not exactly half an hour. It's just that it slows down for some 
time (10, 20, 30 minutes) and then it's OK again. It happens several 
times per day. I checked if there are other processes running when we 
have this slow down but it's not the case.
There's not really a difference between when it's OK or not (apart from 
the number of connections because the db is too slow): load is still at 
4 or 5, iowait is still at 0%, there's still cpu idle and we still have 
free memory. I can't find what is the limit and why there is cpu idle.

I forgot to give our non default postgresql.conf parameters:
shared_buffers = 28800
sort_mem = 32768
vacuum_mem = 32768
max_fsm_pages = 350000
max_fsm_relations = 2000
checkpoint_segments = 16
effective_cache_size = 270000
random_page_cost = 2

Thanks for your help

--
Guillaume