I am following the Memory-Hog thread, and I doubt
that there is an algorithm which satisfies everybody.
But running a compute server for a numerical analysis
group, I am aware that OOM conditions happen (especially
Matlab is a big memory hog). These things, however, always
are exceptional situations and different people would want
different ways to react to such a condition.
I think it would be great if OOM handling would be
configurable (either kernel compile time or /proc interface)
to choose one of the following methods:
a) log it, die and reboot automatically
b) use algorithm XY to shut down the Bad One (like Rik's)
or choose between different algorithms
c) stop the kernel (I know this is non-trivial) and give
somebody (root on the console) the possibility to
choose MANUALLY which processes to kill (preferably with
choice of signal) (BTW: This is probably the method I would
choose, since having a few jobs with runtimes of some days
I don't care that they wait for a few hours until they may
proceed)
Not only for OOM situations but for many exceed limits situations
applicable (open files, ...).
Best regards,
Hermann
-----------------------------------------------
Hermann Schichl
Inst. of Math., Univ. of Vienna, Austria
e-mail: Hermann.Schichl@esi.ac.at
God is real, until declared integer.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/