Torsten Förtsch <torsten.foertsch@gmx.net> wrote:
> Is there an other way to make the planner use generate the 1st
> plan?
The planner cost factors are based on the assumption that a
moderate percentage of random page reads will need to actually go
out to disk. If a high percentage of pages are in cache, you may
want to reduce random_page_cost to something closer to (or even
equal to) seq_page_cost. I generally find I get better plans if I
raise cpu_tuple_cost to 0.03. effective_cache_size should
generally be between 50% and 75% of machine RAM. If these changes
(or others of their ilk) cause costs to be estimated in a way which
more nearly matches reality, better plans will be chosen.
> Why does it generate the 2nd plan at all?
It has the lowest estimated cost, based on your memory
configuration and cost factors.
> Does the planner take into account what is currently present in
> shared memory?
No. If you search the archives you can probably find previous
discussions of whether it would be a good idea to do so; the
consensus has been that it would not be.
If you have further performance-related questions, please review
this page so that you can provide enough information to allow
people to give the most helpful advice:
http://wiki.postgresql.org/wiki/SlowQueryQuestions
--
Kevin Grittner
EDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
