Re: questions about applying DAML-S

>At CMU we are developing a matching algorithm between profiles that
>uses DAML and matches between inputs and outputs (we intend to extend
>on preconditions and effects, but that requires a rule language that
>DAML do not support yet).  When you look for a service, you compile a
>profile of the service you desire, and submit it to a DAML-enabled
>registry.  The registry browses its DB of service advertisements and
>extracts only those services whose inputs match the inputs of your
>request, and whose outputs match the outputs of your request (I'll be
>happy to talk with you about the details of the algorithm, but I will
>skip them for now.)  As far a DAML-enabled registries, we are
>considering services like UDDI but with some semantic capabilities.
>We are currently experimenting with adding such a semantic matching
>capabilities to the UDDI registry by compiling DAML-S advertisements
>in UDDI records and then matching between them.  This is still work in
>progress, but we have some encouraging preliminary results.


I'm very interesting in the algorithm you matching the services. I have 
some comments. Of course, it could be wrong.

The automated retrieval and composition of software components is a 
long-term topic in automated software engineering area. The people use 
automated theorem prover or program synthesis for retrieval and composition 
the software components described by formal languages. This is a good way 
to locate a suitable component from a large repository. However, due to the 
complexity of domain knowledge, no such methods have been used in 
industrial applications. But I think the tide of web services can instigate 
putting those methods into practice, because, from the software engineering 
viewpoint, the web services have no different to the software components 
but the requirement of automated discovery and composition is one of most 
important thing in web service research. Do you agree this opinion?

Comparing with general speaking software components, web services are more 
business related. So we can't discover and composite them only by 
input/output. A workflow model can help a lot on working process and domain 
knowledge. But workflow model has no help on looking for the suitable (or 
the best) service component from the tons of available components having 
the similar function. So, the combination of workflow model and components 
discovery method (reasoning or synthesis) could be a good direction for 
services retrieval and composition. However, the current implementations 
mostly focus on workflow model and work little on matching and discovery 
method. So, I'm very interesting in what you have done.

Jinghai

Received on Thursday, 7 March 2002 16:13:21 UTC