[Gllug] ulimits

[Alex Hudson]

On Sun, Sep 02, 2001 at 03:33:37AM +0100, Jon Masters wrote:
> > it could add it all up and if user 'fred' owns a process over say 40%
> > cpu then that process will get reniced to really priority.
> 
> Ideally, one would also stop processes which went over another
> threshold,

You guys seem to have missed the point. The data gathering is the well easy
part - doing something sensible with it isn't. You can't just stop a
process. It doesn't work like that. Try implementing it on any half-complex
program and watch your machine lock up solid. This would also screw threaded
programs particularly nastily.

Scheduling isn't an easy topic, it's not just about giving processes
runtime. The scheduler is finely tuned and based on well-known (and proven)
algorithms. Unless you do some elementary study of scheduling I can
guarantee anything you do won't get past kernel boot - scheduling is
hardcore compsci.

I also don't see the point of cutting off users programs at arbitrary
limits - what a waste of CPU cycles. If there are cycles free, why not use
them? This is exactly the reason nice is around. Plus, it doesn't break
programs.

If you are actually serious about this area of research you're on the wrong
operating system. Go play on GNU/HURD - your project can be integrating
RT-Mach into oskit. A relevant paper would be Rajkumar (1998). Then you can
go about writing the relevant userland tools and translators required to
make your dream come true.

Cheers,

Alex.

-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug