>>> "Omar Kilani" <omar.kilani@gmail.com> wrote:
> During CS storm:
> count(*) from pg_locks - 1000 - 1400
What do you have for max_connections?
With the hardware and load you describe, I would guess you would limit
context switching and see best performance with a connection pool that
queues requests, keeping the actual connections to the database around
30. You only have so many resources available; having a large number
of queries all contending for them is less efficient than having just
enough queries active to keep them all busy. I have typically seen
the plateau fall off to degradation at or before (CPU count * 2) +
(spindle count).
-Kevin