Re: malloc() patches round 3

[Brent A. Fulgham]

On Fri, Aug 24, 2001 at 10:34:52AM +0200, eric.rannaud.liste@libertysurf.fr 
wrote:
> On Thu, Aug 23, 2001 at 01:22:38PM +0200, a memorable day, Jeroen Dekkers 
> wrote for eternity :
> > What about carefully choose a process and kill that? If for example
> > mozilla starts eating memory, I don't want my system to reboot. I want
> > my system to kill mozilla instead. The process should be chosen carefully,
> > because you don't want an essiental task getting killed. The best way to
> > do it is to choose the process which uses the most memory and isn't one
> > of the essiental servers. Everything else keeps running then and the
> > process which is killed can be restarted if necessary.
>  
> I think we could even avoid killing processes.
> 
> If a sort of DMZ is set for VM, as Igor suggested, we can decide to
> stop (like C-z) the "non-essential processes" (which have to be defined).
> Say that, when e.g. 95% of VM is full, these processes are
> stopped, avoiding more memory consumption, then you have the
> opportunity to control the system.
> Many decisions can be taken here:
> _ killing the responsible process (but that should be avoided as far
> as possible, since data could be lost).
> _ (temporarily) using a new swap partition/file, the user should be
> warned but otherwise I think it could be quite transparent.
> _ (...)
> 
> In all cases, procedures to free memory may be undertaken (as
> evocated before in this thread).
> 

IBM's AIX (Unix Variant) uses a very similar behavior.  And it's quite
handy -- I always suspect a memory leak when I get a report that
a process on one of our client's machines is "hung."  Often AIX will
halt a process once it grows to a configurable limit (set as a boot
parameter.)

I think this would be a nice way for the Hurd to handle things as
well.

-Brent