mlw writes:
> I don't think this is true at all. Making buffers and sort larger numbers, will
> improve performance dramatically.
I think we want to make the buffers bigger by default but we need to put
some thought into the numbers. Currently, the shared buffers are set so
that they presumably fit under the default shared memory limit on most
systems. We already know this isn't actually true anymore.
Tom Lane thinks that "a few thousand" buffers is enough, so let's say 2048
= 16MB. This is certainly a lot less than the 64MB that were proposed or
the 512MB that some people use. However, I feel we should have *some*
data points before we commit to a number that, as you say, most users will
implicitly be stuck with. Even more so if some users claim "the more the
better" and the leading developer in the field disagrees.
(Perhaps we could arrange it that by default the system attempts to
allocate X amount of shared memory and if it fails it tries smaller sizes
until it succeeds (down to a reasonable minimum)? This could combine
convenience and optimimal default.)
As for sort memory, I have no idea why this isn't much larger by default.
> I would also bet that most users NEVER see the postgresql.conf file,
> and just blame poor performance on bad design and start using MySQL.
This could be an information deficit more than anything else. I don't
know.
--
Peter Eisentraut peter_e@gmx.net
