[ILUG] oom fixes and kernel traffic...

[Paul Jakma]

Paul Jakma
Compaq EOMC
Eastpoint, Dublin.

paul.jakma at compaq.com      Tel: +353 1 818 4303
>  That's the side Microsoft took in the Windows 3.11 days - if 
> a program
> crashes your machine, that's not our problem - don't run the program.
> 
>  If your machine is unusable, for anyr reason, what so ever, 
> it's Linux's
> fault.
>  

true... but any OOM situation is going to lead to disaster: either deadlock
or  
processes killed via a generic OOM algorithm. the best would be a tunable
OOM algorithm where you could set policies, etc...

but global resource limits are kernel policies aswell, aren't they?


>  I would be happy to go back to the 1.x days, where the 
> machine would just
> return -ENOMEM to programs that were looking for extra 
> memory, and wouldn't
> swap themselves unusable.
>

i think that behaviour is still there.. somewhere... just don't know where.
The documentation on proc/sys/vm/overcommit_memory isn't very clear. is it 1
to disallow over commit, or the other way round?

 
>  I've seen people with 256MB RAM running out of memory, and 
> halting the
> machine, from netscape over-caching...that's not on. 

i had that aswell. then i set resource limits, and now netscape pop's up a
little box complaining about lack of memory, bash complains it can't fork if
i type a command.. but programmes already running keep running, and other
users' processes are unaffected. and everythings hunky dory.

that seems imo better than: netscape's bloats out memory. processes start
dying (maybe netscape, but maybe not). What happens if X gets killed?

regards,

paul.