On Thu, 13 Jun 1996 Mark.Hemment@uniplex.co.uk wrote:
> >From kmb203@psu.edu
> > One (too) simple choice might be to charge for kernel time .. CPU
> > spent swapping for a process is added to CPU load from that process.
> > If CPU % is then quotad it will keep their process from causing more
> > than it's percent of system load.
>
> I more generic solution would be scheduler classes, as per SVR4 (with
> extensions, of course).
> Trusted users could have their login shell in a class which allows
> high CPU usage - i.e. their processes would be allowed to occupy
> the high level run queues. As child process inherit their scheduler
> class from their parent's, all should be well.
> Other users would be given a class which is only allowed to occupy
> the lower level queues.
> ...other variations...
>
> Scheduling of processes could be based upon real memory usage, page
> faults, etc, as well as CPU usage - depending upon the class.
> Different classes could be written for different environments, with
> the schedulers loaded as modules :)
>
> markhe
>
>