Re: [PATCH] oom killer (Core)

[Andrea Arcangeli <andrea@suse.de>]

On Thu, Dec 02, 2004 at 08:24:10PM +0100, Thomas Gleixner wrote:
> because oom killed sshd. So I cannot send anything except a service man,
> who drives 150km to hit sysrq-F or the reset button.

You'd need a little old computer next to your server with serial console
attached to it for it to be effective. But it sounds more like a last
resort and it makes economical sense only if you're not hosting your
server and you've your own server room. Normally one doesn't need to
drive 150km just because you can call somebody at the phone to click
reboot (worse than SYSRQ+F [unless it was the critical app itself
hitting a memleak], but much cheaper than hosting a serial console
server too).

The SYSRQ+F sounds more useful on a desktop usage, if you've tons and
tons of swap. Don't forget that with an huge amount of swap, you're
telling the kernel "please make my machine extremely slow if all apps
uses the ram at the same time". In many situations it may be more
efficicient for you to kill one of these apps, wait the other to finish,
and restart the killed app from scratch (to avoid trashing, and the
kernel _can't_ avoid trashing or it wouldn't be fair anymore). So if you
do a mistake and you want to recover responsiveness in less than a
second, the SYSRQ+F trick should make it.