Steve,
> We're splitting the front-end across a number of machines, but all of
> the various datasets are sufficiently intertwined that they all have to be
> in the same database. I'm going to fiddle around with some of the
> available replication options and see if they're robust enough to put them
> into production.
2 other suggestions:
1. Both PostgreSQL Inc. and Command Prompt Inc. have some sort of pay-for HA
solution for Postgres. Paying them may end up being cheaper than
improvising this yourself.
2. Zapatec Inc. has acheived impressive performance gains by putting the
database on a high-speed, HA gigabit NAS server and having a few "client
servers" handle incoming queries. You may want to experiment along these
lines.
--
-Josh Berkus
Aglio Database Solutions
San Francisco