Re: DAML-S ws composition: Is the cycle complete?

On Monday, September 29, 2003, at 04:37  PM, Charlie Abela wrote:
> Hi all,
>
> my questions are simple and hopefully relevant.
>
> How important for the WS composition cycle is the caching of newly 
> created WS
> descriptions (especially the process model)?

Thus far, not very? :) I'd suspect that it's never going to be *very* 
important, except, *maybe* as a performance boost (and, then, I'd look 
for some internal datastructure, not a DAML-S description, but ok, I 
can maybe see something). If we're talking a planning system, then 
remember that you aren't likely to get anything fundamentally or 
dramatically *new* out of this sort of "bootstrapping".

> I am asking this question because I haven't seen work/discussions that 
> gives it
> the due importance, so is it because this is not an issue or what?

I don't know of anyone that uses such inside a "composition cycle" so 
it's currently moot. Between such cycles, you can just use the normal 
discovery mechanisms.

> Has anyone done work on composition where the output is a DAML-S 
> description
> that goes beyond the use of sequences? I know about the work from the 
> mindswap
> and UMBC groups, but to my knowledge none of these have transformed 
> their plans
> into complete DAML-S definitions which make use of other control 
> constructs (in
> the process model).

We've just started extending our SHOP2 work to generate conditional 
plans (basically involves reading off the task network for a successful 
plan). We've also been working own extending our interactive composer 
to handle control constructs (simpliciy, dataflow, control flow...pick 
1 :)).

Again, I'd be very surprised if the conditional plans themselves made 
very interesting methods. They might, as I said, tune performance 
(though I'm skeptical).

Perhaps you could explain what you were expecting, and how you think it 
might be achieved?

Cheers,
Bijan Parsia.

Received on Monday, 29 September 2003 23:13:57 UTC