Re: questions about applying DAML-S

Paul,

Two points to follow up on Massimo's response to your question.

Don't forget that the service profile is the advertisement for the
service.  In part, it is populated from the process model, which
describes the actual program that realizes the service.  Hence,
inputs, outputs, preconditions and effects come from the process model
if one exists.

As for how to handle a "failure", it depends on what you mean by
failure.  If you mean a network error, system down, program execution
error (e.g., variable out of bounds) then currently we don't model 
these in the service profile or process model. Some of this will be modeled
in the process execution model, which we are working on now, but is
not part of version 0.6 of DAML-S.

Alternately, if by "failure" you mean that the web service executed
successfully but did not have the "intended outcome"  (e.g., you
were trying to buy a book at amazon.com, but the book was out of stock)
then we can handle these "failures" and they are simply effects 
that are conditional on state (e.g., the book not being carried or
out of stock).

Sheila


Massimo Paolucci wrote:

>  >
>  > 3. We are having trouble deciding whether a "failure" is an output or an
>  > effect.
>  > The DAML-S tech overview says
>  > "Effects are events that are caused by the successful execution of a
>  > service."
>  > The walkthrough says
>  > "Note that not all services have physical side-effects.
>  > In particular, services that are strictly information providing do not."
>  > So if AeroDAML fails to return a markup file is this an effect?
>  > Or is some form of error message an output?
>  > What about if you get no error message and something just hangs?
> 
> I am still unsure about the role of failures in the service profile
> and I expect that it will depend on how it will be used.  One of the
> purposes of the service profile is service location, from that point
> of view you want to search on the bases of what the service does, not
> how the service fails.  Still, the service profile can also be used in
> other ways (such as decide which service to use when there are ties
> being one) and there the failures may become important.  We did not
> address this problem in the service profile yet.  The process model of
> DAML-S does contain the information on how to detect failures, so
> services can look at what kind of failures may occur while interacting
> with the service.

==============================================================================

Sheila McIlraith, PhD                 Phone: 650-723-7932
Research Scientist                    Fax:  650-725-5850
Knowledge Systems Lab
Department of Computer Science
Gates Sciences Building, 2A-248       http://www.ksl.stanford.edu/people/sam
Stanford University                   E-mail sam@ksl.stanford.edu
Stanford, CA 94305-9020

Received on Thursday, 7 March 2002 15:09:36 UTC