Re: [GENERAL] Upgrade to dual processor machine? - Mailing list pgsql-performance

From Andrew Sullivan
Subject Re: [GENERAL] Upgrade to dual processor machine?
Date
Msg-id 20021115093459.C19503@mail.libertyrms.com
Whole thread Raw
In response to Re: [GENERAL] Upgrade to dual processor machine?  (Cedric Dufour (Cogito Ergo Soft)<cedric.dufour@cogito-ergo-soft.com>)
List pgsql-performance
On Fri, Nov 15, 2002 at 02:37:01PM +0100, Cedric Dufour (Cogito Ergo Soft) wrote:

> The 'Runtime configuration / General operation' part of the doc is quite
> short on the subject.

I'm afriad the setting was new in 7.2, and people don't have a great
deal of experience with it.  So it's difficult to ake
recommendations.

> Is there any other places to look for more details on this FSM ? What policy
> should drive changes to the FSM settings ?

If your tables change a lot between VACUUM, the FSM may fill up.  the
problem is that the system is only to keep "in mind" so much
information about how many pages can be freed.  This affects the
re-use of disk space.

> I guess allowing larger FSM values might improve UPDATE performance (require
> VACUUM less often) but consume RAM that may be more useful elsewhere. Am I
> right ?

Not really.  Your better bet is to perform VACUUM often; but if you
don't do that, then VACUUM will be able to re-claim more space in the
table if your FSM is larger.  Is that clear-ish?

You can estimate the correct value, apparently, by doing some
calculations about disk space and turnover.  I think it requires some
cycles where you do VACUUM FULL.  There was a discussion on the
-general list about it some months ago, IIRC.

A
--
----
Andrew Sullivan                         204-4141 Yonge Street
Liberty RMS                           Toronto, Ontario Canada
<andrew@libertyrms.info>                              M2P 2A8
                                         +1 416 646 3304 x110


pgsql-performance by date:

Previous
From: Richard Huxton
Date:
Subject: Re: for/loop performance in plpgsql ?
Next
From: Tom Lane
Date:
Subject: Re: Sort time