From: | "Scott Otis" <scott(dot)otis(at)intand(dot)com> |
---|---|
To: | "Ivan Voras" <ivoras(at)freebsd(dot)org>, <pgsql-performance(at)postgresql(dot)org> |
Subject: | Re: Seeking performance advice and explanation for high I/O on 8.3 |
Date: | 2009-09-03 20:16:30 |
Message-ID: | F406B740738CB44A8BCBEF9734ECF663683F36@albus.intand.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
Sorry about not responding to the whole list earlier - this is my first time posting to a mailing list.
Would providing more information about the size and complexities of the databases help?
I measure I/O stats with iostat - here is the command I use:
iostat -d -x mfid0 -t 290 2
I tried looking at the man page for iostat but couldn't find anywhere how to determine what the stats are for sequential vs random - any help there?
When using 'top -m io' the postgres stats collector process is constantly at 99% - 100%.
When using 'top' the WCPU for the postgres stats collector and the autovacuum process are constantly at 20% - 21%.
Is that normal? It seems to me that the stats collector is doing all the I/O (which would mean the stats collector is doing 46.1 megabytes /sec).
Also, the I/O stats don't change hardly at all (except at night during backups which makes sense). They don't go up or down with user activity on the server - which makes me wonder a little bit. I have a feeling that if I just turned off Apache that the I/O stats wouldn't change. Which leads me to believe that the I/O is not query related - its stats collecting and autovacuuming related. Is that expected?
It seems to me that the stats collector shouldn't be using that much I/O and CPU (and the autovacuum shouldn't be using that much CPU) - therefore something in my configuration must be messed up or could be changed somehow. But maybe I'm wrong - please let me know.
I don't think my setup is necessarily slow. I just want to make it as efficient as possible and wanted to get some feedback to see if am setting things up right. I am also looking out into the future and seeing how much load I can put on this server before getting another one. If I can reduce the I/O and CPU that the stats collector and autovacuum are using without losing any functionality then I can put more load on the server.
Again thanks for all the help.
Scott Otis
CIO / Lead Developer
Intand
www.intand.com
From | Date | Subject | |
---|---|---|---|
Next Message | Andy Colson | 2009-09-03 21:09:08 | Re: Seeking performance advice and explanation for high I/O on 8.3 |
Previous Message | astro77 | 2009-09-03 20:06:26 | Re: Slow select times on select with xpath |