Re: questions about disk configurations - Mailing list pgsql-performance

From Adrian 'Dagurashibanipal' von Bidder
Subject Re: questions about disk configurations
Date
Msg-id 1039435317.819.36.camel@altfrangg.fortytwo.ch
Whole thread Raw
In response to questions about disk configurations  (Hubert depesz Lubaczewski <depesz@depesz.pl>)
Responses Re: questions about disk configurations
List pgsql-performance
On Mon, 2002-12-09 at 12:32, Hubert depesz Lubaczewski wrote:
> hi
> i have a question about best harddisk configuration for postgresql
> performance.
[...]

Yo!

A bit more data is needed before anybody can give you more help:
 - what is your budget?
 - how big will your databases be?
 - what's the read/write ratio?

Even then you'll not get any good recipes, because there aren't any.
You'll have to do benchmarks yourself. A few fundamental things that are
probably true for most:

 - more RAM is always good. Independent from the disc architecture - if
an access isn't going to the disc at all, it's always good. (if you're
mostly writing this may be lessened).
 - always carefully tune the postgres installation (random page cost,
sort mem, shared buffers, ... - all depend on your system and you
application)
 - as you correctly said: distribute the load on many spindles. On a
busy database, 4*20G is probably faster than 1*80G

beyound this, experiences vary. RAID1 and RAID5 are rated differently by
different people - and especially with RAID5 there are (I think) really
performance differencies between the various products. RAID0 is fastest,
of course, but you probably care for your data.

For equally good implementations, RAID1 and RAID5 may have similar
speed, especially if the RAID controller for RAID5 has enough RAM. If
the active dataset on a RAID5 is bigger than the available caching RAM,
write performance sucks as a single block write requires 2 reads and 2
writes. If the RAID5 controller has enough RAM (and a decent
implementation), write performance can be almost equal to RAID1 (2
writes for a single block write).

So far
-- vbi


--
this email is protected by a digital signature:  http://fortytwo.ch/gpg

NOTE: keyserver bugs! get my key here: https://fortytwo.ch/gpg/92082481

Attachment

pgsql-performance by date:

Previous
From: Hubert depesz Lubaczewski
Date:
Subject: questions about disk configurations
Next
From: Rod Taylor
Date:
Subject: Re: is insertion and movement times are correlated to