> > I'll present a way to make 72 completely unusable:
> >
> > Take stress, run stress mega mega mega .... (many times). Then
> > enjoy. (I assume that number of mega must be bigger than your RAM+swap
> > ammount).
> >
> > Machine is not dead, interrupt still works, magic unmount works, maybe
> > it is even some userland working - kswapd is active 'only' 99% of
> > time. (Debugging leds are _great_ think), hard disk does somesing very
> > short about once from two seconds or so. I can see led of user
> > processes flagshing _very_ shortly.
>
> I think the system should kill some (large) non-root processes
> when (free_mem + free_swap) gets below (free_pages_low +
> min_free_pages)/2 and stays there for several (5?) seconds.
> That kill should (ofcourse) be logged to syslog.
> If killing all non-root processes doesn't work, just kill _all_
> processes, remount fs-es RO, try to umount() as many as possible
> and reboot with a message...
Sorry. There is a problem. It needs to be solved, not worked
around. (Notice, that same process does nothing bad to 2.0.28).
And: Work around is bad. Imagine your machine with such behaviour on
100MBit ethernet. Imagine me around (ping -f)ing your machine. That
can keep your pages low for as long as I want. You do not your machine
to go yo-yo (up and down and up and down ...).
Pavel
-- I'm really pavel@atrey.karlin.mff.cuni.cz. Pavel Look at http://atrey.karlin.mff.cuni.cz/~pavel/ ;-).