[ILUG] Out of Memory

[Paul Jakma]

On Tue, 7 Mar 2000, John P. Looney wrote:


>  There is also that Linux locks up, completely, while Solaris/FreeBSD
> don't. And if they are cleverer at hosing userspace, doesn't that make
> them better at avoiding userspace ?
> 

well, i suppose i could claim on a technicality that linux hasn't locked
up completely. It's just deadlocked, if there was some way to free
memory it would continue. But then i guess i'd be evading your main
point.


>  But it is. That's why they are fixing it!
> 

it's not the VM they are fixing. It's the process killing stuff they're
fixing. Plus the ENOMEM thing. (but Stephen Tweedie reckons not
returning ENOMEM on OOM might actually be a good thing.. so maybe that's
not a bug).

>  The tmpfs is a bit dogdy. You are using swap as /tmp. Writing into it,
> and filling it up should make the machine run out of RAM. However, most of
> the cases that I've seen that happen, the machine is still usable to some
> degree (at least enough to be able to telnet in & wipe it).
> 

true. Like i said already, to sol's credit, the admin had no problem
telneting in, and he cleared tempfs and i could continue working
straight away. That was nice.


>  But you can't. What about a rogue process that could be buggy, that a
> developer could be working on, and it goes nuts. It happened me loads of
> times about a year ago.
> 

did you have limits set?

i've had exactly one hang on linux because of OOM. On a box that's used
as nfs/mail/dial-up server + X desktop/apps via XDM/telnet. I had
reinstalled RH, and forgot to add pam_limits.so to the pam.d/kdm file
(on my old RH setup i had pam.d/gdm pam.d/kdm linked to pam.d/xdm)..

only time it's ever happened to me. How can linux behave correctly for
me, incorrectly for you if linux is so broken?

>  2.2.x is a lot better than 2.0.x, but can still be hard-hung.
> 

true. but by the same measure you can prevent OOM with limits...

> Kate
> 
> 

-paul.