RE: requirements summary

Jean-Jaques
 
I think the problem you mention is solved as long as you assume that
error/fault messages are included in the choreogrpahy. For example, if Party
B can reply to A with either: a "normal" (ie. non-error) message, OR an
error/fault message; then if C sends a fault to B, then B can send a fault
to A. The two interactions are decoupled.
 
Another alternative to sending a fault message is to allow A to inquire on
the state of B.  If B has failed then A will discover this from B's
response.
 
I think that a choreography that does not include appropriate methods of
allowing one party to discover the state of another choreography will
normally be a badly designed choreography ;). The only exception that I can
think of is a choreography that involves a one-way "fire and forget" message
where the sender does not care what happened to it - but then one message is
hardly a choreography!
 
Thoughts?
 
David

-----Original Message-----
From: Jean-Jacques Dubray [mailto:jjd@eigner.com]
Sent: Tuesday, March 25, 2003 9:26 AM
To: 'Burdett, David'; 'Jean-Jacques Dubray'; 'Ricky Ho'; jdart@tibco.com;
Daniel_Austin@grainger.com
Cc: public-ws-chor@w3.org
Subject: RE: requirements summary



David: 

thanks, 

At the run-time engine level, things gets far more complicated because 
unless there is a party that touches all the "bilateral choreographies", 
it is impossible without special run-time to "monitor" the multi-party 
choreography. So the question arise, is the goal of a multi-party 
choreography specification to allow configuration of run-time engines? 
<DB>It depends what you mean by "Monitor". Each party can monitor their own
behavior and the behaviors of the other roles with which they interact. If
one of the parties discovers that some other party is not behaving properly,
then they can raise errors with that party.</DB>

[JJ] Let's take a more concrete example, such as the propagation of
exceptions, if a failure happens (e.g. an operation returned a fault), to
between party B and C. How do we notify party A? Are we expecting
choreography designers to explicitly define the corresponding message
exchange between B and A should this happen? Or are we expecting a more
generic mechanism by which A can be notified of the corresponding "state" of
the choreography. This could be implemented by the run-time infrastructure.
Of course that complicates quite a bit this implementation.

In think in the light of this, we should not conclude that binary is a 
special case of multi-party. They may well have both distinct features 
(control flow?) and applications. 
<DB>I'm not sure there is difference, but let's keep exploring ;) </DB> 

[JJ] This is more a question ;-) than an assertion.

I am also wondering if the group wants to keep as a requirement that 
says that in the choreography specification there is no distinction 
between the choreography involving "internal" services as opposed to 
external services. A separate layer of the specification should allow 
for annotating that this particular message exchange is external and may 
have more qualifiers. However, at the pure choreography specification 
level, the choreographies should not be distinguished. 
<DB>Am I right in assuming that by "internal" you mean within a "domain of
control", e.g. a business, and that "External" means between domains of
control, e.g. between businesses. If so then although, in theory, they can
be expressed in the same way, there are significant *practical* differences:

1. External choreographies, between domains of control, will be used by
MULTIPLE (perhaps millions) of different parties and therefore the
definition MUST be abstract to avoid multiple definitions of essentially the
same choreography.

2. For Internal choreographies, within a domain of control, there is only
ONE implementation and therefore the definition can be very concrete and
does not need to be abstract at all.

[JJ] I agree that this is generally true but large companies might also want
to benefit from this abstraction. We have customers that have 30 ERP
implementations, I also often talk about this large A&D company that has 84
procurement systems. 

JJ-  
  

>>-----Original Message----- 
>>From: public-ws-chor-request@w3.org 
[ mailto:public-ws-chor-request@w3.org
<mailto:public-ws-chor-request@w3.org> ] 
>>On Behalf Of Ricky Ho 
>>Sent: Monday, March 24, 2003 7:06 PM 
>>To: jdart@tibco.com; Daniel_Austin@grainger.com 
>>Cc: public-ws-chor@w3.org 
>>Subject: Re: requirements summary 
>> 
>> 
>>I was originally thinking that a multi-party choreography can always 
be 
>>broken down into multiple "inter-dependent" bi-party choreography. 
But I 
>>am convinced that this is NOT always possible. 
>> 
>>See 
http://lists.w3.org/Archives/Public/public-ws-chor/2003Mar/0052.html
<http://lists.w3.org/Archives/Public/public-ws-chor/2003Mar/0052.html>  
>> 
>>So I think bi-party choreography is a special case of multi-party 
>>choreography.  Bi-party choreography has some interesting properties 
that 
>>can simplify the modeling.  (e.g. Bi-Party choreography doesn't need 
to 
>>worry about dynamic participation because any change of a binding can 
>>simply terminate the choreography). 
>> 
>>I think we should covered multi-party choreography.  In additional, we 
may 
>>also need to investigate this special subset called bi-party 
choreography. 
>> 
>>Best regards, 
>>Ricky 
>> 
>>At 02:28 PM 3/24/2003 -0800, Jon Dart wrote: 
>> 
>>>Daniel_Austin@grainger.com wrote: 
>>>>2. Multi-party vs. bilateral choreography: there is some skepticism 
>>>>that modelling bilateral interactions is sufficient. 
>>>>       I certainly don't think that is it sufficient to model only 
>>bilateral 
>>>>transactions. Many business transactions have multiple actors, and 
we 
>>want 
>>>>to build standards that will work for common service transaction 
models. 
>>> 
>>>Note that it is not exactly all or nothing here. BPSS for example 
>>supports 
>>>"MultiParty Collaborations", but does so by composing them out of 
"Binary 
>>>Collaborations". 
>>> 
>>>--Jon 
>>> 
>>>