Dag-Erling Smorgrav wrote:
>
> "Daniel C. Sobral" <d...@newsguy.com> writes:
> > * a sysctl to make the system non-overcommit
>
> So I see common sense lost in the end.
I think nobody objects to the knob, just to people trying to
convince us that it would do any good.
> > * SIGDANGER in low-memory situations
>
> Do we support more than 32 signals?
So it's a cascade project. :-)
> ISTR AIX already does this. What signal numbers / names does AIX use
> for this?
It's AIX that I have in mind when I think of this. (AIX does have
the knob, which can be set per process.)
> > * Dividing processes into those that ought to be killed first and
> > those that ought to be killed last in low-memory situations
>
> How does AIX solve that problem?
AFAIK, it doesn't. Though maybe the processes which are not
overcommitting are "immune", which makes some sense.
> > * Per-user swap space limit
>
> Is that a realistic goal? What do we do about shared text, count it
> once for every user that uses it?
Shared TEXT is not swapped. :-) We are talking about *swap*. I don't
think we could have any swap space shared between users except
through some arcane uses of mmap(). Anyway, DG suggested, Dillon
thought it a good idea, who am I to say anything?
--
Daniel C. Sobral (8-DCS)
d...@newsguy.com
d...@freebsd.org
"Misguided Angel hanging over me
Heart like Gabriel, pure and white as ivory
Soul like Lucifer, black and cold like a piece of lead..."
To Unsubscribe: send mail to majord...@freebsd.org
with "unsubscribe freebsd-hackers" in the body of the message
