Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!julius.cs.uiuc.edu!apple!agate!shelby!rutgers!rochester!pt.cs.cmu.edu!o.gp.cs.cmu.edu!andrew.cmu.edu!<UNAUTHENTICATED>+
From: David.Maynard@CS.CMU.EDU
Newsgroups: comp.realtime
Subject: Re: Software primitives for real-time programming languages
Message-ID: <Ab0upj600ju9IPTkoa@cs.cmu.edu>
Date: 28 Sep 90 20:07:11 GMT
Organization: Carnegie Mellon, Pittsburgh, PA
Lines: 53
In-Reply-To: <1892@tuvie>

> >to plan beforehand and say: under the following conditions, we know
> >that tasks A, B, and R will be preempted indefinately, while tasks
> >X and Y are still guaranteed to make their deadlines, which they MUST
> >in every circumstance.

> I can imagine that value functions could be used for this purpose as
> well.

To date, our research has focused on dynamic systems that are
characteristic of supervisory real-time control.  We feel that these
types of guarantees are almost always inappropriate for that level of
systems.  Some types of guarantees can simplify failure recovery.  The
main problem is whether you can REALLY make the guarantees. 
Statically-designed systems that have been well-planned and that operate
in a sufficiently controlled environment are good examples of when the
approach can work well.   However, as applications become more complex
and demanding it is becoming increasingly difficult to build these types
of systems.  In more dynamic systems the "guarantees" that are made
often don't really buy you anything (and may be counterproductive).

> So in order to be able to employ those in safety-critical systems
> one would have to have a-priori quantitative measures for the
> reliability of
> such systems .. i.e. statements like `under the stated circumstances the
> probability that this tasks misses its deadline is less than 1e-6'....

This would be one approach.  We haven't (yet) considered it.

> But I believe that the techniques necessary for such `best-effort'
> systems 
> are not necessarily the same than those for hard real-time systems.

I've only recently started thinking about how we could address low-end
control systems in the same environment as the higher-level functions. 
That is one reason I'm particularly interested in learning what
specification primitives people think are appropriate.  I don't
(necessarily) expect the same techniques will work for both domains.  I
do want to want the approaches to be compatible.

> Of course I go along with David Maynard in hoping for a unified
> theory and method for building both types of systems. 
> But this is going to be a very difficult goal, to say the least.

I agree.

 ---
 David P. Maynard (dpm@cs.cmu.edu)
 Alpha OS Research Group
 Carnegie Mellon University & Concurrent Computer Corp.
 ---
 Any opinions expressed are mine.  I haven't asked CMU
 or Concurrent what they think.
 ---