Tom Lane wrote:
> When benchmarking with data sets considerably larger than available
> buffer cache, I rather doubt that small random_page_cost would be a
> good idea. Still, you might as well experiment to see.
From experience, I know the difference in response time can be huge when postgres incorrectly
chooses a sequential scan over an index scan. In practice, do people experience as great a
difference when postgres incorrectly chooses an index scan over a sequential scan? My intuition
is that the speed difference is a lot less for incorrectly choosing an index scan. If this is the
case, it would be safer to chose a small value for random_page_cost.
George Essig