Re: On PATH_MAX

[ness]

Jonathan S. Shapiro wrote:
> On Thu, 2005-11-10 at 21:01 +0100, ness wrote:
>
> > Jonathan S. Shapiro wrote:
> >
> > > On Thu, 2005-11-10 at 20:00 +0100, ness wrote:
> > >
> > >
> > > > > The client cannot grant a guarantee that the client does not have. The
> > > > > client's schedule is revocable, therefore any schedule that it might
> > > > > grant is revocable.
> > > >
> > > > Why should a given schedule be revokable?
> > >
> > >
> > > If you cannot revoke schedule, how do you correct an erroneously issued
> > > schedule?
> >
> > I'm not an expert and I have not yet read anything about reservation 
> > based scheduling, but let me raise the question: why should a schedule 
> > be issued erroneously?
>
>
> Because I type "1ms in 50ms" when I meant "2ms in 100ms", or because I
> start a program with a schedule and later realize that I do not want it
> to run.

Hm. So the implications are as follows:
There are operations in exectution that must be finished, and they must 
be finished in a reasonable time. They might operate on memory, so this 
memory has to be available for a reasonable time (the time the operation 
needs). We want the caller to pay for what he wants. So at least parts 
of the memory and cpu time given by him have to be unrevokable. So the 
caller needs the ability to provide unrevokable memory and cpu time 
(well, at least both caller and callee have to trust the source of the 
memory/cpu time not to revoke it; of course unrevokablility might be 
testable). This means, things like kill -9 cannot work.
How can this be solved (regardless to cost-value ratio)?
Maybye additionally to standard schedules very less non-revokable 
schedules could be issued. This wouldn't make kill -9 work, but if it 
are few enough non-revokable it would act mostly like one expected it in 
a system where the caller doesn't pay. Of course it would be hard to 
find the amount of non-revokable schedules to be issued, or only a 
reasonable default.
But my proposal would add even more complexity, and I dunno whether one 
should call it rsonable, as tools like kill -9 would work like before...
--
-ness-