> I don't think random_page_cost actually corresponds with any real number
> anymore. I just treat it as an uncalibrated knob you can turn and
> benchmark the results at.
And, frankly, not a useful knob. You get much more useful results out
of effective_cache_size and cpu_* costs than you get out of messing with
random_page_cost, unless you're running on SSD or something which would
justify a lower RPC, or if you're compensating for our poor n-distinct
estimation for very large tables.
--Josh