Re: How to get RTREE performance from GIST index?

On Sun, Nov 22, 2009 at 10:55:21AM +0000, Clive Page wrote:
>> You are AFAICR the first person to have a problem is this area, but if
>
> I find it extremely hard to believe that. All I am doing is finding
> whether pairs of rectangular boxes overlap or not. That is the most
> trivial use of R-trees possible. Surely someone thought to time that
> using GIST?

Ofcourse, there were benchmarks, which showed the GiST version to be
faster or the same.

http://www.sai.msu.su/~megera/postgres/gist/code/rtree/README.rtree_gist

Not surprising really, since conceptually the GiST version resembles the
old rtree code very closely, which is why people are surprised you're
seeing a difference. Hence people as suspecting that the problem lies
elsewhere.

(GiST is basically the extension of rtree of non-geometric types, there
really isn't that much difference between the two).

> I agree that this is a bug in Postgres - the bug was removing code that
> worked perfectly well and upon which some users depended. I simply
> don't understand why the Rtree code could not have been left in there,
> for those who found that the new-fangled GIST indexing did not work.

I think there would be some disagreements about whether the old code
was "working well", it was broken enough that people didn't want to
maintain it.

Looking forward to your explain output.

Have a nice day,
--
Martijn van Oosterhout <kleptog(at)svana(dot)org> http://svana.org/kleptog/
> Please line up in a tree and maintain the heap invariant while
> boarding. Thank you for flying nlogn airlines.