Re: rtree indexes aren't being used with 7.0

At 11:13 AM 15-05-2000 -0400, Tom Lane wrote:
>Jeff Hoffmann <jeff(at)propertykey(dot)com> writes:
>Actually, the cost estimator for indexscans is deliberately not linear,
>since it's trying to model the effects of hits in a buffer cache ...
>perhaps that's a pointless refinement when we have no accurate idea of
>the size of the kernel's buffer cache, but certainly the real-world
>behavior is not going to be linear.

Just have something like a few linear "curves" and some IF statements?
That's often good enough in real world scenarios- the whole point about
fuzzy logic. Works very well for high speed lifts, and other seemingly
complex control situations.

Cheerio,
Link.