What everybody else said. :) But if it comes to voting...
Anything to improve parallelism is good.
Anything reducing blocking (ie: CLUSTER, VACUUM FULL) is good
Improved handling of sort_mem (I think this will hit bizgres first)
merge :)
STATISTICS ON INDEXES! (specifically multi-field indexes)
Multiple query plans for bound parameters.
Materialized views. I don't know the history behind why Slony is
trigger-based, but I think both materialized views and replication would
benefit greatly from having a means to tie into WAL (or something
similar) instead of using triggers. I would expect this to result in a
dramatic speed improvement over triggers, since you would no longer be
double-logging. A slick way to do this would be to tie-in to WAL writes
that meet certain criteria (namely that they hit a specified table) and
store those seperately on-disk. These would be played-back as needed.
This mechanism should be useful for both replication and MViews. If you
look at one of Oracle's replicaiton options, it's actually just a form
of MViews that are on remote machines. Even if we stick with something
trigger-based for now I think we should provide a base mechanism that
works for both MViews and replication.
--
Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com
Pervasive Software http://pervasive.com 512-569-9461