Re: Non-linear Performance

Curt Sampson <cjs@cynic.net> writes:
> What is the advantage, if any, to having postgres do the buffering
> in its shared memory rather than letting the OS do it?

Not much, if any.  I don't believe in making shared_buffers be more than
(at most) 25% of physical RAM.  In most cases it's just as effective to
keep it smaller.  I would recommend bumping up the default though ;-).
Something in the low thousands (of buffers) is probably a realistic
minimum.

            regards, tom lane