RE: [ebsoa] Does SOA Require Registry-Based Dynamic Discovery?

[James Bryce Clark <jamie.clark@oasis-open.org>]

    In five years I still haven't found an occasion to
disagree with anything Yunker says.
    Personal opinion only, purely FYI:

    1.  Few I've heard would limit "SOA" to
be limited, necessarily, only to  infinitely extensible
architectures that permit unanticipated newcomers to access all
services.  
    2.  Counterexample:  Some "SOA"
may be purely inside a secure  firewall, or close trading community,
where all consumers know about all services on offer (or else they're not
eligible to get 'em).  Functionally, one might imagine sneaker-net
exchange of WSDLs, or ignoring WSDL altogether when no
"description" is needed because there is a steady state of
finite, known services.  
    3.  People are doing that today.  There is a
British e-Science community who attends GGF, composed mostly of
physicists and radiotelescope jockeys, as an example.  Were you to
ask them, they would say "yes, we are using web
services".  But you might not see much in their messages that
you recognize other than SOAP.  In some cases maybe not even
that.  The phrase "I don't see why I need all this bloody
stuff" has been uttered more than once in their sessions. 
    One could get into debates about the definitions, and
which tribal boundaries or best practices are infringed by that behavior,
but they are, in fact, sharing and exchanging production scientific data
using these methods.  
    4.  What John describes as an "open"
SOA is one, I think, which has that quality of facilitating unanticipated
newcomer consumers and service offers, without the n(n-1) problem. 
The added feature is, functionally, an interrogatable external
representation of the services in a shared space.  A number of
architects and standards organizations would like to canonicalize that
resource directory space:  UDDI, ebXML Registry, tote board screens
at Caesar's Palace, cave walls at Lascaux, whatever.
    5.  To my view, CPPA mechanics are one method for
placing virtuous constraints on such an "open" system, by
better defining the who and when and with-what of
service availability.  Without which participants cannot meet their
business needs.  Other projects (such as the WS-Addressing activity
at W3C, seeking to make WSDL useful and more meaningful;  and the
work on statefulness from our WSDM TC) also pursue that goal.  I am
curious -- but underinformed -- about which of those various constraint
methods will be composable. 
    Regards JBC