Re: Finding bottleneck - Mailing list pgsql-performance

From Tom Lane
Subject Re: Finding bottleneck
Date
Msg-id 8434.1123529245@sss.pgh.pa.us
Whole thread Raw
In response to Re: Finding bottleneck  (Kari Lavikka <tuner@bdb.fi>)
List pgsql-performance
Kari Lavikka <tuner@bdb.fi> writes:
> Disk configurations looks something like this:
>    sda: data (10 spindles, raid10)
>    sdb: xlog & clog (2 spindles, raid1)
>    sdc: os and other stuff

That's definitely wrong.  Put clog on the data disk.  The entire point
of giving xlog its own spindle is that you don't ever want the disk
heads moving off the current xlog file.  I'm not sure how much this is
hurting you, given that clog is relatively low volume, but if you're
going to go to the trouble of putting xlog on a separate spindle then
it should be a completely dedicated spindle.

            regards, tom lane

pgsql-performance by date:

Previous
From: Kari Lavikka
Date:
Subject: Re: Finding bottleneck
Next
From: Tom Lane
Date:
Subject: Re: Why hash join instead of nested loop?