Re: Out of memory kernel death

Samuli Kaski (samkaski@cs.Helsinki.FI)
Thu, 8 May 1997 11:54:31 +0300 (EET DST)

On Wed, 7 May 1997, James Mastros wrote:

> Wouldn't killing off the process that has the lowest CPU use be best
> (other than processes 1-10, which are reserved, and currently all not
> terminatable without dire consiqunces), on the theory that it wasn't doing
> anything anyway? That should atleast give you enough memory to kill -HUP
> others instead of just destroying them.

I think this has been discussed before and all the methods brought
up were found to be inadequate for common usage. Someone was always
able to find a situation in which the algorithm in question didn't
work as it was supposed to.

But anyway I will throw in my input:

Another method is to use the living time of processes. Usually
system processes (including netscape and x :)) live for a
considerable amount of time growing slowly in memory allocation
and cpu usage. Normally all the forking/bombing/malfunctioning
programs take up considerable amounts of resources immediatly.
Ofcourse if such an feature is implemented there has to be a way
to disable it for some processes since we still want to be able
to run qw clients and servers :)

(not forgetting those linux boxes that do number crunching
and non-desktop work)

What am I not seeing here?

Samuli Kaski,
Department of Computer Science, University of Helsinki, Finland.