Re: [moca] all-you-can-eat calculi

[sanjiva@xxxxxxxxxxxxxxxxx]

I forgot to add this to my last response:  If one has a synch broadcast
implementation, why can't one implement the "all-you-can-eat" calculus
by the dualizing translation where every send becomes
a receive and every receive becomes a send?  Is there some asymmetry
in your notion of observation that this translation is not meaning
preserving?



> thanks to all who replied to my query. some further comments below.
>
> martin
>
> -------- From: Uwe Nestmann --------
>
>  > MB> i call it "all-you-can-eat" -- where one input consumes
>  > MB> as many outputs as possible in one go.
>  >
>  > Do you mean "as many as available" or "as many as wanted"?
>
> the former, but i suspect that both share features.
>
>  > MB>     x.P | \overline{x} | ... \overline{x}  --->  P
>  >
>  > "As many as available" is not expressed by this rule alone,
>  > because it does not enforce it, but only enables it.
>
> that's true. i didn't want to burden my post with too much formal
> detail. in addition, "As many as available" can be achieved by various
> formal means, none of which has emerged as a clear favourite.
>
>  > Apart from this, you may want to look into the (early?)
>  > works on LOTOS,
>
> thanks, i shall have a look.
>
> -------- From: Pawel Wojciechowski --------
>
>  >  > one input consumes as many outputs as possible in one go.
>  >
>  > why only one, not all inputs on 'x' ?  :-)
>
> that's an interesting question. the application i have in mind needs
> exactly one input on x interacting with arbitrarily many outputs on
> that channel. i use a typing system to models this, starting from an
> untyped calculus that allows to have many inputs on x in
> parallel. there are various ways one can set up the semantics for
> multiple parallel inputs, but they all resolve to the same behaviour
> for the restriction to unique inputs.
>
>  > not sure if useful (probably not) but perhaps you might
>  > think of distributed algorithms with unknown number
>  > of processes (that can dynamically join/leave/crash/
>  > migrate in ad-hoc networks etc.)
>  >
>  > e.g. a distributed consensus algorithm accepts a number
>  > of outputs (each with a possibly different value) and
>  > completes by reducing some number of inputs (each input
>  > consumes exactly the _same_ value, chosen from those that
>  > were output).
>
> thanks, this is very interesting. so you are saying, each process
> nondeterministically chooses to output a number on x. if all numbers
> coincide the interaction happens between all outputting participants
> and the receiver(s)? i see a little problem with this: either
>
>   * the set of numbers to be used is finite, in which case the
>     algorithm has a scalability limit and is not compositional, or
>
>   * it is unlimited. then all participants may agree on a number that
>     does not correspond to an existing process, which means the
>     election needs to be redone. but then it may fail again etc leading
>     to the possibility of divergence.
>
>
> -------- From: GOUAICH Abdelkader --------
>
>  > Can't you already express this by the (infinite) recursion operator?
>
> I'm not sure how to do this in a compositional way
>
>
>
>
>
>
>
>
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> The "models for mobility" mailing list     mailto:moca@xxxxxxxxxxxxxxx
>  http://www-sop.inria.fr/mimosa/personnel/Davide.Sangiorgi/moca.html
>

  
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The "models for mobility" mailing list     mailto:moca@xxxxxxxxxxxxxxx
 http://www-sop.inria.fr/mimosa/personnel/Davide.Sangiorgi/moca.html