On Fri, Mar 23, 2007 at 02:13:57PM +0100, Henrik Zagerholm wrote:
> I'm just amazed that setting the statistics threshold on one column
> mad all the difference. IS there any guidelines on what columns I
> should change the statistics on?
Start by looking for columns involved in simple comparisons with a
constant -- especially if you know that the distribution of values
is uneven -- and see how far off the row count estimate is for
various values. You can use simple queries such as
EXPLAIN ANALYZE SELECT 1 FROM tbl_file WHERE fk_filetype_id = 83;
To learn more read "How the Planner Uses Statistics" in the
documentation:
http://www.postgresql.org/docs/8.2/interactive/planner-stats-details.html
Among the configuration settings to consider changing are shared_buffers,
effective_cache_size, work_mem, and random_page_cost. The following
is a good starting point:
http://www.powerpostgresql.com/PerfList
If you have additional performance-related questions then consider
posting to pgsql-performance.
--
Michael Fuhr