Re: In the belly of the beast (MySQLCon) - Mailing list pgsql-general

From Scott Marlowe
Subject Re: In the belly of the beast (MySQLCon)
Date
Msg-id dcc563d10804201032v6f955965l791a425f7ffae2d6@mail.gmail.com
Whole thread Raw
In response to Re: In the belly of the beast (MySQLCon)  (Scott Ribe <scott_ribe@killerbytes.com>)
Responses Re: In the belly of the beast (MySQLCon)  ("Joshua D. Drake" <jd@commandprompt.com>)
Re: In the belly of the beast (MySQLCon)  (Csaba Nagy <nagy@ecircle-ag.com>)
Re: In the belly of the beast (MySQLCon)  (Greg Smith <gsmith@gregsmith.com>)
List pgsql-general
On Sun, Apr 20, 2008 at 11:12 AM, Scott Ribe <scott_ribe@killerbytes.com> wrote:
> > I am going to play with this and see where it breaks, but it's going to be
>  > an enormous time investment to babysit it.
>
>  One thing to remember, since you've already got backup in place and this
>  replica would just be for upgrading, not a hot spare, turn off fsync during
>  the initial subscription. My data set is quite a bit smaller, and I've
>  gotten into the habit of turning off fsync during the initial post-upgrade
>  load, to shorten my downtime.

Exactly.  There are several things you can do on the replica that you
wouldn't do on the master for better performance.  Really big numbers
of WAL segments, no background writer, fsync off as mentioned, WAL on
a RAID-0 with 10 disks, no battery back on a caching RAID controller,
and so on.  You also might look into cheap but useful things like
bonded gigabit networking between the two servers with a dedicated
switch / rolled cable.

I wonder if there's a comprehensive list somewhere...

What I keep dreaming of is a process that lets slony use pg_bulkloader
or something like it to do the initial load...

pgsql-general by date:

Previous
From: Scott Ribe
Date:
Subject: Re: In the belly of the beast (MySQLCon)
Next
From: Peter Eisentraut
Date:
Subject: Re: How to tell if 64 bit vs 32 bit engine?