Re: go for a script! / ex: PostgreSQL vs. MySQL - Mailing list pgsql-performance

From Josh Berkus
Subject Re: go for a script! / ex: PostgreSQL vs. MySQL
Date
Msg-id 200310141009.26247.josh@agliodbs.com
Whole thread Raw
In response to Re: go for a script! / ex: PostgreSQL vs. MySQL  ("Ivan Voras" <ivoras@fer.hr>)
Responses Re: go for a script! / ex: PostgreSQL vs. MySQL  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-performance
Ivan,

> There was some talk, either on this list or freebsd-performance about
> setting the default block size for PostgreSQL running on FreeBSD to be 16k
> because of performance reasons. That is: *default* for the port, user is
> not asked. So an automagical method to scale non-default block sizes is a
> very needed thing.

Hmmm ... possibly.   My concern is that if someone uses a very non-default
value, such as 256K, then they are probably better off doing their own tuning
because they've got an unusual system.  However, we could easily limit it to
the range of 4K to 32K.

Of course, since there's no GUC var, we'd have to ask the user to confirm
their block size.  I'm reluctant to take this approach because if the user
gets it wrong, then the settings will be *way* off ... and possibly cause
PostgreSQL to be unrunnable or have "out of memory" crashes.

Unless there's a way to find it in the compiled source?

> > 2) may not work well for anyone using unusual locales, optimization
> > flags, or other non-default compile options except for language
> > interfaces.
>
> Depends on what you consider 'unusual'? I hope not things like iso8859-x
> (or, to be exact, European languages) :)

On second thought, I'm not sure what an "unusual locale" would be.  Scratch
that.

--
Josh Berkus
Aglio Database Solutions
San Francisco

pgsql-performance by date:

Previous
From: Tom Lane
Date:
Subject: Re: [HACKERS] Sun performance - Major discovery!
Next
From: Josh Berkus
Date:
Subject: Re: Large Text Search Help