Re: pg_trgm performance

Hi Steinar,

On 2/24/07, Steinar H. Gunderson <sgunderson(at)bigfoot(dot)com> wrote:
> I'm sorry, I can no longer remember where I needed pg_trgm. Simple testing of
> your patch seems to indicate that the GiN version is about 65% _slower_ (18ms
> vs. 30ms) for a test data set I found lying around, but I remember that on
> the data set I needed it, the GIST version was a lot slower than that (think
> 3-400ms). The 18 vs. 30ms test is a random Amarok database, on 8.2.3
> (Debian).

Could you post EXPLAIN ANALYZE for both queries (after 2 or 3 runs)?
And if you can provide EXPLAIN ANALYZE for a couple of searches
(short length, medium length and long) in both cases, it could be nice
too.

The GiN version is not selective enough currently compared to GiST. It
generally finds the matching rows faster but it has a slower recheck
cond so it's sometimes interesting (in my case) and sometimes not that
interesting (it seems to be your case).

Thanks.

--
Guillaume