> > >These are all minor abberations though, on the whole the estimates
>are
> > >pretty good. Perhaps you need to tweak the values of random_page_cost
> > >and similar variables.
> >
> > Thank You, It's general problem or only mine? I have "100%"
> > standard current PC.
>
>The default random_page_cost assumes some concurrent activity. If your
>PC does nothing else concurrently, the performance of a seq scan will
>be underestimated.
>
>Try to do the statement with some concurrent disk load and you will most
>likely
>see that the 1. plan is faster. (assuming the tables are not fully
>cached)
>
>Andreas
ok. I tested it with pgbench and it's true. With -c 50 merge_join is
faster. I didn't expect it.
Thank You
Pavel
_________________________________________________________________
Citite se osamele? Poznejte nekoho vyjmecneho diky Match.com.
http://www.msn.cz/