Hi,
I've implemented Bob Jenkin's hash function for PostgreSQL; more
information on the hash function can be found at
http://burtleburtle.net/bob/hash/doobs.html
I'm posting this to -hackers because I'd like to get some feedback on
whether this actually improves performance. I've tested 2 situations
locally:
(1) pgbench, btree indexes (to test typical performance)
(2) pgbench, hash indexes
I haven't looked at the implementation of hash joins; if they happen to
use this hash function as well, that would be another informative
situation to benchmark.
In my local tests, the performance in (1) is basically the same, while
the performance in (2) is increased by 4% to 8%. If you could let me
know the results on your local setup, it should become clear whether
this patch is a performance win overall.
Note that to test case (2) properly, you'll need to drop and re-create
your pgbench indexes after you apply the patch (so that when the new
indexes are created, they use the new hash function). Also, it would be
a good idea to use concurrency level 1; at higher concurrency levels,
hash indexes have a tendancy to deadlock (yes, I'm currently trying to
fix that too ;-) ).
Cheers,
Neil
--
Neil Conway <neilconway@rogers.com>
PGP Key ID: DB3C29FC