Re: mosbench revisited

Jeff Janes <jeff.janes@gmail.com> writes:
> My experiments have shown that the freelist proper is not
> substantially faster than the freelist clocksweep--and that is even
> under the assumption that putting things back into the freelist is
> absolutely free.

The freelist isn't there to make buffer allocation faster, though;
it's there for allocation efficiency.  The point is that when some
buffers have become completely useless (eg, because we dropped the table
they were for), they'll be recycled in preference to reclaiming buffers
that contain still-possibly-useful data.  It would certainly be simple
to get rid of the freelist and only recycle dead buffers when the clock
sweep reaches them, but I think we'd be paying for that in extra,
unnecessary I/O.
        regards, tom lane