To: Manfred Koizar <mkoi-pg@aon.at>
Cc: pgsql-hackers@postgresql.org
Subject: Re: Correlation in cost_index() 
In-reply-to: <n6cmpug13b9rk1srebjvhphg0lm8dou1kn@4ax.com> 
References: <n6cmpug13b9rk1srebjvhphg0lm8dou1kn@4ax.com>
Comments: In-reply-to Manfred Koizar <mkoi-pg@aon.at>
	message dated "Wed, 02 Oct 2002 21:52:46 +0200"
Date: Wed, 02 Oct 2002 18:48:49 -0400
Message-ID: <29475.1033598929@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Manfred Koizar <mkoi-pg@aon.at> writes:
> AFAICS (part of) the real problem is in costsize.c:cost_index() where
> IO_cost is calculated from min_IO_cost, pages_fetched,
> random_page_cost, and indexCorrelation.  The current implementation
> uses indexCorrelation^2 to interpolate between min_IO_cost and
> max_IO_cost, which IMHO gives results that are too close to
> max_IO_cost.

The indexCorrelation^2 algorithm was only a quick hack with no theory
behind it :-(.  I've wanted to find some better method to put in there,
but have not had any time to research the problem.

> As nobody knows how each of these proposals performs in real life
> under different conditions, I suggest to leave the current
> implementation in, add all three algorithms, and supply a GUC variable
> to select a cost function.

I don't think it's really a good idea to expect users to pick among
multiple cost functions that *all* have no guiding theory behind them.
I'd prefer to see us find a better cost function and use it.  Has anyone
trawled the database literature on the subject?

			regards, tom lane