From: | Andres Freund <andres(at)anarazel(dot)de> |
---|---|
To: | Cédric Villemain <cedric(dot)villemain(dot)debian(at)gmail(dot)com> |
Cc: | pgsql-performance(at)postgresql(dot)org, Michael Kohl <michael(dot)kohl(at)tupalo(dot)com> |
Subject: | Re: High load, |
Date: | 2011-01-27 13:26:38 |
Message-ID: | 201101271426.38876.andres@anarazel.de |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-performance |
On Thursday, January 27, 2011 02:23:48 PM Cédric Villemain wrote:
> 2011/1/27 Andres Freund <andres(at)anarazel(dot)de>:
> > On Thursday, January 27, 2011 12:24:10 PM Cédric Villemain wrote:
> >> > maintenance_work_mem = 512MB
> >>
> >> 128MB is usualy enough
> >
> > Uhm, I don't want to be picky, but thats not really my experience. Sorts
> > for index creation are highly dependent on a high m_w_m. Quite regularly
> > I find the existing 1GB limit a probleme here...
>
> That is right for index creation, but not for 'pure' maintenance
> stuff. Once the database is running as usual, there is no really point
> to give auto-vacuum or auto-analyze much more (depend on the raid card
> memory too ...)
Even that I cannot agree with, sorry ;-). If you have a database with much
churn a high m_w_m helps to avoid multiple scans during vacuum of the database
because the amount of dead tuples doesn't fit m_w_m.
Andres
From | Date | Subject | |
---|---|---|---|
Next Message | David Greco | 2011-01-27 13:48:00 | Re: Real vs Int performance |
Previous Message | Cédric Villemain | 2011-01-27 13:23:48 | Re: High load, |