On Wed, Oct 9, 2013 at 10:45:52AM -0400, Bruce Momjian wrote: > On Wed, Oct 9, 2013 at 04:40:38PM +0200, Pavel Stehule wrote: > > Effectively, if every session uses one full work_mem, you end up with > > total work_mem usage equal to shared_buffers. > > > > We can try a different algorithm to scale up work_mem, but it seems wise > > to auto-scale it up to some extent based on shared_buffers. > > > > > > In my experience a optimal value of work_mem depends on data and load, so I > > prefer a work_mem as independent parameter. > > But it still is an independent parameter. I am just changing the default. > > > maintenance_work_mem can depend on work_mem ~ work_mem * 1 * max_connection / 4 > > That is kind of hard to do because we would have to figure out if the > old maintenance_work_mem was set from a default computation or by the > user.
FYI, this auto-tuning is not for us, who understand the parameters and how they interact, but for the 90% of our users who would benefit from better defaults. It is true that there might now be cases where you would need to _reduce_ work_mem from its default, but I think the new computed default will be better for most users.
then we should to use as base a how much dedicated RAM is for PG - not shared buffers.
