On Thu, Feb 03, 2011 at 05:53:28PM +0200, Peter Eisentraut wrote:
> On tor, 2011-02-03 at 00:10 -0500, Noah Misch wrote:
> > > This is good stuff. I'll send you a new patch in a day or three for
> > > perhaps another round of performance tests. Some of the other
> > issues
> > > above can perhaps be postponed for follow-up patches.
> >
> > I agree -- if the performance-when-unused gets solid, none of my other
> > comments ought to hold things up.
>
> Here is a new patch.
>
> The main change is in varstr_cmp(), avoiding the calls to
> pg_newlocale_from_collation() when the default locale is used. This
> accounts for the performance regression in my tests. It also addresses
> some of your refactoring ideas.
Looks good and tests well. I've attached the same benchmark script with updated
timings, and I've marked the patch Ready for Committer.
nm
-- Setup
-- SELECT setseed(0);
-- CREATE TABLE t (c) AS
-- SELECT chr(1 + (random() * 65534)::int) FROM generate_series(1,10000000)
gen(n);
-- NOTE: due to concurrent load, timings typically varied 1-2s between runs.
-- en_US.UTF8, unpatched: 68.75s
-- en_US.UTF8, patched: 67.95s
-- id_ID.UTF8, unpatched: 67.48s
-- id_ID.UTF8, patched: 69.34s
SELECT min(c)
-- en_US.UTF8, patched: 81.59s
-- id_ID.UTF8, patched: 84.59s
--SELECT min(c COLLATE "id_ID")
FROM (
SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
UNION ALL SELECT c FROM t
) t_all;
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
