Re: Words for the Triangles

Warning: long tirade follows:

Lets be clear here on what it is we are defining.  We are defining a Web
Services Architetcture.  It is a case of a Service Oriented Architecture.
There are 3 primary parties in this model.  I'm not suggesting it. I'm
stating it.

We have had SOAs before, as Anne has pointed out... both Corba and RMI use
registries, but you had to know what you wanted and what his name was and
where its registry was before you could get access to it.

I think Web services adds to this the ability to find what you want by
SEARCHING on characteristics and metadata accompanying that service.  This
is SUCH a good thing for J2EE.

Here is my experience: the web was not interesting to me or the general
populace until yahoo, google, and dogpile (my favorite search engine
aggregator). Why?  Because THEN I could FIND stuff by some search criteria
or taxonomy.  Before that, there's a 99% chance I don't have the URL for
what I want or even KNOW who it is that has it.   So what works on the
web????? FINDING from some CENTRAL thing... and of course there can be many
many (and there are) central things (also known to some as registries)!
You have to ask SOMEBODY to find the data!

Now, how did that central thing get it entries?  They either got registered
there manually (Yahoo, and MANY others) or the registry crawled the web
(google and many others) using meta description data at the URL and content
processing to create its own entries.  (Oh, I know, we don't CALL these
registries, we call them search engines... but what are they searching???
the entire web for every query??? Ha! They are searching their own stash of
URLs..... a registry!) All this applies equally to web services.  Someone
can register the URL of the WSDL and/or service to a registry with meta
data - UDDI-ish, or it can crawl looking for WSDLs or WSILs at URLs and
register the services therein.  UDDI plays the same role as Yahoo for web

<tirade>Speaking of which it mystifies me why UDDI triggers all these 'bad'
vibes because its a 'central registry'... there's certainly NOT only one,
there will be many and they may federate with each other, or not.  Yet WSIL
is not percieved as a 'central registry', but, it is within the scope of
its host. Any list of entities is a 'central' registry for some scope or
domain.  In either case its a list of WSDL URLs for a domain. There's no
requirement that they be only in one registry/wsil. The main differentiator
I see operationally is that WSIL is not searchable and doesn't have higher
level of descriptions factored in (taxonomy, metadata, business, etc), and
has a predicable location. UDDI is searchable and entries can be more fully
described with business, taxonomy, etc.  They are both very usefull for
their scenarios and essential to the Web services story. If I were writing
a crawling web services registry (woogle?) I'd certainly be looking for
WSILs and WSDLs at URLs, I'd also be crawling and harvesting UDDI
registries I find.  And I cannot imagine why we won't see the emergence of
'more searchable' service registries in the future that can search on the
details of WSDL documents.</tirade>

So, I think that a 'registry' role plays a first class part in a web
services architeture... just like it does for 'why the web works'.

Now, just because it place a first class role doesn't mean that its
required for every single scenario... just like the web. I fully agree and
in the words that I've submitted I say that the publish/find operations are
very broadly defined to 'any way you get a wsdl to someone else' and 'any
way you get a wsdl'... and sometimes a registry is used to facilitate that.
Just like the web where sometimes you get the URL from someone directly,
sometimes from searching some registry.  Sometimes you get a WSDL directly,
sometimes you get it through seaching some registry.

Either way, we cannot be dismissive of the importance and role of
registries in the web, or web servics.

So 'somehow you get a WSDL' isn't crisp enough.
Its not magic happens here... its what we're here to define.

Oh boy, I can see the flames lighting up already...

Heather Kreger
Web Services Lead Architect
STSM, SWG Emerging Technology
919-543-3211 (t/l 441)  cell:919-496-9572

Mark Baker <> on 09/26/2002 12:35:47 PM

Sent by:

To:    Ugo Corda <>
cc:    "'Hugo Haas'" <>,
Subject:    Re: Words for the Triangles


Maybe so, but it still suggests the three-party model; A wants to
communicate with B, but is required to go to C in order to get the
necessary information to do so.

We know how to enable communication without a registry.  It's not
difficult.  Let's promote that.



On Thu, Sep 26, 2002 at 08:57:42AM -0700, Ugo Corda wrote:
> >I think that UDDI hints at a central registry solution, and putting it
> >in a sentence such as "the key to reaching this new horizon is a
> >common program-to-program communication model" definitely pushes in
> >this direction.
> Version 3 of UDDI has moved away from the concept of a central registry.
> UDDI 3 supports multiregistry topologies (which is different than version
> 2's multinode topologies based on node replication). For more details,
> UDDI 3 section 8, "Publishing Across Multiple Registries" [1].
> Ugo
> [1]

Mark Baker, CTO, Idokorro Mobile (formerly Planetfred)
Ottawa, Ontario, CANADA.     

Received on Thursday, 26 September 2002 14:08:57 UTC