Re: Position paper

[Jonathan S. Shapiro]

On Tue, 2007-01-09 at 23:45 +0100, Pierre THIERRY wrote: 
> Scribit Tom Bachmann dies 06/01/2007 hora 19:26:
> > I don't think cpu time pools are to be passed to servers. Although
> > this would increase accounting, it would as well horrify the
> > complexity of the server and require special kernel support, as has
> > been discussed on the list (or on coyotos-dev?).
> 
> Won't Coyotos has scheduler activations, or the primitives needed to
> implement them? I thought they would make such designs relatively easy
> to implement? (but I confess I did not dig the issue of scheduler
> activations deeply, and I barely grasp the concept...)

The two issues are orthogonal. Scheduler activations is simply a
different interface between the kernel-level dispatcher and the process.
Time pools have to do with when a process is elligible to run.

The hard problems to solve with time pools are

  a) Efficiency of the mechanism

  b) Defining what happens when a server is running on a donated
     time pool and the time runs out.

Issue (b) is particularly important. You CANNOT simply stop execution.
There MUST be some way for the service to re-establish a sane state.

In EROS, we concluded that this was all too hard, and that if you needed
an end to end schedule the right thing to do was establish an end to end
arrangement of private service threads.

shap