Re: Performance question (stripped down the problem)

On Thu, Sep 27, 2001 at 11:18:31AM -0600, Steve Wolfe wrote:
> This is interesting, just yesterday I was perusing some of Bruce
> Momjian's works on PG tuning, and noticed that Postgres prefers sequential
> scans over indexes when much of the table has to be read, all because of
> the number of head movements on the disk. It would seem that these days,
> where RAM is cheap, that most people have a great enough disk cache that
> head movements can become irrelevant.
>
> However, I can also see where some people may have incredibly large
> tables that just won't fit into RAM. An easy solution to both might be to
> create a user-specifiable switch passed at startup that would simply tell
> PG that sequentials aren't necessarily better than index scans. Not
> completely disabling them, but at least giving it a pointer that it
> doesn't *have* to use sequentials.

There is a user specifieable value somewhere that controls how expensive an
index scan is and how expensive a seqential scan is. By tuning those you
could probably get the effect you want.

--
Martijn van Oosterhout <kleptog(at)svana(dot)org>
http://svana.org/kleptog/
> Magnetism, electricity and motion are like a three-for-two special offer:
> if you have two of them, the third one comes free.