Re: WS-Policy and SAWSDL

On Sep 13, 2006, at 11:09 AM, Jacek Kopecky wrote:

> Hi Bijan,
>
> personally, I would say the difference between SAWSDL and Policy is
> this: SAWSDL gives information about what the annotated WSDL does,

You mean the service?

> on a
> higher level than the name of the interface. Policy gives information
> about how (and under what conditions) it does it, assuming the client
> already knows what it does.

Not a sharp line, obviously. But in any case, both are critical for  
discovery.

> SAWSDL annotations are meant to help the client (human or automaton)
> find the service that Does What The Client Wants(tm).

The Way That the Client Wants To Do It(tm).

> Policy is meant to help the client that already knows a service (or a
> set of them) to check that the service can do it the way the client
> wants.

Well, first I don't see SAWSDL is so restricted. You can express QoS,  
security, and similar information. Part of the problem is that SAWSDL  
is very vague on what it wants.

Seems like you are claiming that it's like UDDI (particularly,  
categories). But obvoiusly, we need *all* the information, function  
signature, "what it does", other constraints to do discovery. Hence  
my feeling weirded out by your making the discovery argument.

> If the client wants to get current stock information, it wouldn't use
> Policy to select from among billing, shipping, weather and stock
> services - here's where it would use SAWSDL.

SWASDL is just an attachment mechanism, right? It doesn't actually  
offer any expressivity in the representation. So I don't see how you  
can claim this. SAWSDL is competing with Policy *attachment*, not  
policy.

> If the client has a number of stock information services, it will use
> policies to select the one that will give most up-to-date (QoS)

I fail to see why anyone would think you needed distinct mechanisms  
for this.

> information for acceptable price, over an encrypted channel. SAWSDL
> wouldn't help here.
>
> BTW, both tasks above fall under matchmaking as various people would
> define it, yet still I see them as two different tasks.

That require entirely different, unrelated syntax? That's what you  
have to justify, eh?

> That's what I call difference of intent.

But there has to be good evidence that difference of intent should be  
reflected in the language in this way. I don't see any argument for  
that. The age analogy certainly fails.

> Of course the nature of
> semantic annotations and policy assertions is that nobody wants to
> constrain them, but the tools people are likely to build with  
> SAWSDL or
> WS-Policy will, by market forces, only do some tasks with the
> information. One can use Policy as a framework for a programming
> language, yet nobody will build a commercial system around that (as  
> far
> as I can foresee).

Not an effective reductio. The purposes are much more closely aligned  
and the expressive capacity of the associated representation  
languages too similar. It's not like one bends either beyond it's  
normal ken.

Indeed, you keep sliding to the representation language and  
ontologies, which SAWSDL doesn't constrain or influence at all, afaict.

> It is possible that in the future the industry will stop being  
> afraid of
> semantics and automation, and semantics people will start caring about
> QoS parameters, and that QNames and URIs will be reconciled (haha),  
> and
> a single unified framework for both policies and semantic annotations
> will be created. I won't fight against this, but I don't see the will
> currently to go there, yet there is will to have the partial solutions
> standardized now.

I fail to see why this matters for an attachment mechanism.

> So from a practical, fairly-short-term point of view, the  
> intentions are
> really different.

But I fail to see why this has any technical impact. Why should I use  
sawsdl instead of policy with custom assertion sets?

> Especially in the Web Services activity, the W3C needs
> to have a practical, fairly-short-term point of view,

It is from that point of view that I argue. I don't see the need for  
two standards that do essentially the same thing, esp. when one has  
massive support and the other almost no support.

> and evolution of
> standards is not so expensive.

Yes, the transition from RDF M&S to RDF Core was so cheap and simple.

(Sarcasm :))

I can't get people to agree to change rdf where it's trivially easy  
to fix (e.g., RDF/XML doesn't serialize all legal graphs; a fix is  
easy).

> (see
> http://www.jacek.cz/blog/archives/000075.html )

I still fail to see that doing parallel work helps at all. I've not  
yet seen a single technical reason. I've seen you give some socio- 
political reasons, but they all seem to work *against* SAWSDL.

Now the political reason that there were enough members who want to  
do things that way and they convinced the team is fine. I  
acknowledged that and I'll certainly not argue against it. But that's  
very different that there being a clear, sound technical rationale. I  
still don't see *that*. I only see the opposite (i.e., too many ways  
to say roughly the same thing; no indication of when to use which;  
etc. etc. etc.)

> Hope this clarifies my position,

Well, it's better than the intent for discovery rationale :)

Cheers,
Bijan.

Received on Wednesday, 13 September 2006 11:21:44 UTC