Nailah Ogeer <ogeer@cs.queensu.ca> writes:
> I am currently undergoing the project of making postgres's single buffer
> pool into multiple buffer pools hopefully to decrease the number of
> physical reads. Just wondering if anyone has done some work on this
> before.
To be honest, I think this is a complete waste of effort, as the optimal
buffering scheme for Postgres relies on OS-level caching more than
PG-internal caching. But yes, it's been discussed before. See the
archives for pgsql-performance and pgsql-hackers.
> Also, I need some information on where the number of relations in
> the system is defined.
There is no very practical way to determine that. Any one backend can
only see one database, not the entire cluster; and at the time shared
memory is sized, you don't have any database access capability at all.
regards, tom lane
