Thread: random_page_cost

How should the random_page_cost be calculated?

If our disk system does sustained sequential reads at 400MBps, a
sequential page fetch will take 0.02ms.

If the read seek time is 4ms, does this mean that the random_page_cost
is 200 (4/0.02)?

Or should we use the track-to-track seek time of 0.6ms, which would give
a random_page_cost of 30?

Or something else?

Barry Geipel
(barry@GeipelNet.com)

On Wed, 14 Apr 2004 06:22:56 -0700, "Barry L. Geipel"
<barry@GeipelNet.com> wrote:
>How should the random_page_cost be calculated?

There are so many factors involved that I'd say it cannot be calculated.
The best strategy is to start with the default value of 4 and only fine
tune it based on the run times of real world queries.

If you experience problems with certain queries, post a short
description of your problem and EXPLAIN ANALYSE output to the
-performance mailing list.

Servus
 Manfred