Re: Process instantiation puzzle

On Friday, September 19, 2003, at 05:23  PM, Drew McDermott wrote:

>    [Bijan Parsia]
>    An aside: I conclude from Drew's lack of response that I can declare
>    victory :)
>
> A more plausible conclusion is that I'm drowning in e-mail.

Hmm. But where's the parity? (Or parody?) You get to assign views to 
*me* :)

>   I have
> had an amazing number of offers to share some money left by retired
> (and coincidentally deceased) African dictators, and I expect to
> become very wealthy in the near future.

I hope I'm invited to your "I'm so rich now" party.

>    [Sudhir Agarwal]
>> in my opinion, it depends on, who is controlling the execution of C.
>
>    I disagree. It depends on 1) what services the process occurances
>    actually get instantiated to, and 2) how the service provider
>    implements the advertized semantics.
>
>    Someone, somewhere, has to care about 1. Someone, sometimes, other 
> than
>    the provider, might care about 2 but it may typically be the case 
> that
>    she can't know anything and needn't be bothered by the fact.
>
> Sorry I echoed your somewhere, somewhen, someone prosody.  The puzzle
> seems to cry out for it.

Hmm. Perhaps I've lost track. Where did you do this?

>    I am, of course, assuming that the semantics of these various
>    alternative Drew puts forth all respect the semantics of the
>    description under which P was selected. Indeed, if P was abstract, I
>    can see P's occurences being bound to services from *two different
>    providers*.
>
> I think we're getting somewhere.  What do you mean by "the semantics
> of the description under which P was selected"?

I mean, the IOPEs, plus what's in the WSDL. That's pretty much all I 
have to go on. And from a DAML-S perspective, i.e., for the kind of 
*reasoning* DAML-S explicitly encourages, I think I'm supposed to 
ignore things like whether I use the HTTP or the SMTP binding.

>  Is this description
> rich enough to settle the issue of what sorts of things P1 and P2 are?

Doubt it. Is anything? If I have some degree of glass box description 
of P, I might know more, but if they are just AtomicProcesses, then 
pretty much all I know about them are the IOPEs, plus some endpoint 
stuff.

> Where is the description to be found?  What language is it expressed
> in?  What is the semantics of that language?

I take that my above answers cover this.

>    Take a simpler example. I do a series of GETs on an URI. I have NO 
> CLUE
>    whether I'm getting a single web server, something out of a farm,
>    something from an intermediate cache or proxy. Nor do I care. Nor
>    should I, in most cases.
>
>    I certianly can *build* some of these distinctions in, perhaps, but 
> the
>    possibility of don't know, don't care needs to be preserved, and is
>    probably the default.
>
> I'm willing to let stateless interaction be the default, as for the
> global atomic-clock service.  I want to know how we indicate that
> we've strayed from the default.

We are indeed getting somewhere. I take it that, currently at least, 
the only way we have is via Effects.

That's not quite right, I suppose. We can also know something about 
what services have been invoked (had, or will have, messages sent to) 
by looking at the grounding (though, the grounding doesn't actually 
tell us which binding or endpoint we've used, I believe).

Cheers,
Bijan Parsia.

Received on Friday, 19 September 2003 21:34:04 UTC