> these values [uvmexp.freemin estimation and clamping] actually come
> from 4.4BSD/machvm completely untouched.

But our usage of these resources has changed over time as we have
poolified more data structures...

> many many years ago i spent some time trying to determine better valu=
es
> for this but i didn't really discover anything conclusive...
>
> however, this shouldn't really affect the pool allocator... we alread=
y
> reserve a bunch of pages for such conditions i thought.

What if there is a greedy pool memory user around at the time when we
cross into consuming these "precious/reserve" pool memory resources?
I might very well imagine that the vnode pool can exhibit this sort of
behaviour during the first "daily" run which includes a "find / ...".

I wonder, does the vnode pool code deal well with being told by the
pool allocator "no memory available"?  Should the vnode pool and
possibly other greedy pool memory users which can deal with not being
able to allocate more pool memory resources be prevented from
consuming the pages from this "precious/reserve" area of memory?

Regards,

- H=E5vard