RE: [soa-rm] Opacity calumny

[Ken Laskey <klaskey@mitre.org>]

ramblings on loose coupling:

- I should be able to find a published interface and make use of the 
service solely on the basis of the published information.
- But this sounds more like a documentation issue because I can have a 
600-page document specifying the details of an interface (requiring lots of 
effort to use) or two sentences for a simple interface but it is impossible 
to use because I really needed two pages of documentation.
- Loose coupling seems related to how much effort there is in changing the 
service you use for a given purpose.  That is certainly the effect we are 
after.
- To some extent, it appears loose coupling has to do with the reusability 
of interfaces but that seems to depend on a strategy for versioning (as a 
given service evolves and some of those changes are reflected in the 
interface) and semantic negotiation (because I don't want to rephrase my 
question, i.e. rewrite my interface, for every service providing a similar 
function).
- Profiles are often seen as a solution to the previous item but profiles 
are often static, controlled vocabularies and they run into all those known 
problems.  You will have loose coupling as long as the vocabulary does not 
have to change and you can support still support service evolution.  When 
you want to shift to a capability outside that vocabulary, the coupling has 
suddenly become very tight.
- Another problem with profiles is these are often focused on "business 
verticals" rather than general capabilities.  Thus we create SOA stovepipes 
across which it is difficult to share information.  Again, you may have 
loose coupling within the vertical but tight coupling in the context of 
inter-domain sharing.
- The "perfect" solution is I describe my problem in my vocabulary and the 
service magic translates my information as needed to first discover and 
then assemble and finally use resources to solve my problem.  An 
intermediate phase of the solution may be a message back to me asking for 
additional information, but this information request will either be in my 
vocabulary or, if there is no available translation, referenced to the 
vocabulary of the resource needing the info.  Note here I've mixed 
"resource" and "service" and have not been careful about which one I really 
mean in any given place.  However, I still content I use resources to build 
my solutions and services provide the (hopefully loose coupling) access to 
the resources.

Ken

At 09:14 AM 7/13/2005, Bashioum,Christopher D wrote:
>Agreed - so is there something about the loose coupling and the design
>of the interfaces that is peculiarly SOA (and thus would fit in the RM),
>or is a tightly coupled interface still ok?  I tend to think that we
>need to capture something in the RM that focuses on loose coupling -
>which is similar to the OO concept of public vs. private (Frank's
>earlier assertions).
>
>-----Original Message-----
>From: Michael Stiefel [mailto:development@reliablesoftware.com]
>Sent: Wednesday, July 13, 2005 9:01 AM
>To: Bashioum,Christopher D; soa-rm@lists.oasis-open.org
>Subject: RE: [soa-rm] Opacity calumny
>
>I agree.
>
>Access through interface does not, however, automatically get you loose
>coupling. Poorly designed interfaces can produce tight coupling.
>
>Michael
>
>At 09:37 PM 7/12/2005, Bashioum,Christopher D wrote:
> >  Ken, et. Al.,
> >
> >I was reading your email thread and started thinking about opacity as
>it
> >relates to the concept of loose coupling.  It seems to me that the real
> >issue is loose coupling as opposed to opacity per se.  One could
>publish
> >all the specs related to a service, including the actual code, as part
> >of the service description.  As long as access to the internals of the
> >service is strictly via the interface, then you still have loose
> >coupling (and, a service).  If, on the other hand, you don't tell
>anyone
> >about the internals of a service, but you still allow access to the
> >internals via globals or back-door methods, you will not have loose
> >coupling.
> >
> >I think opacity is really a surrogate for loose coupling.  In other
> >words, its not about what you know of the service via its metadata, its
> >what you can access of the internals of the service (thus the
> >private/public that Frank mentions).  Ken - I think this is also what
> >you were trying to say also (thus, the ever more violent agreements
> >below).  So, that being said, we should probably address this more
> >succinctly in the document.
> >
> >-----Original Message-----
> >From: Ken Laskey [mailto:klaskey@mitre.org]
> >Sent: Monday, July 11, 2005 5:24 PM
> >To: Michael Stiefel; Frank McCabe
> >Cc: soa-rm@lists.oasis-open.org
> >Subject: Re: [soa-rm] Opacity calumny
> >
> >Agree, agree, agree.  This list is getting much too tame.
> >
> >I would suggest that the concept of opacity is relevant to the RM, and
> >the
> >fact that there are tradeoffs and examples of tradeoffs is relevant to
> >the
> >RM.  I also think relating this to a/c/p is relevant to the RM.  Being
> >an
> >engineer, it is difficult to list all the problems and not propose
> >solutions, but alas that is not part of the RM.  So what of this is
> >appropriate for the RA?
> >
> >Ken
> >
> >At 05:09 PM 7/11/2005, Michael Stiefel wrote:
> > >I agree as well.
> > >
> > >Opacity is a design tradeoff and hence a concrete, not abstract
>issue.
> >All
> > >other things being equal more opacity is better than less, but in
> >reality
> > >you are trading off opacity for other things such as how much
>metadata
> >you
> > >publish and how much dependency you generate as a result.
> > >
> > >Michael
> > >
> > >At 04:31 PM 7/11/2005, Ken Laskey wrote:
> > >>Frank,
> > >>
> > >>To agree even more violently, this raises (and I allude to a little
>in
> >
> > >>the draft) the question of how to provide a means to publish
>metadata
> >as
> > >>it is found to be needed but in a way that supports a consistent use
> >of
> > >>the metadata in evaluating an entity against unambiguously stated
> >a/c/p,
> > >>which themselves can be augmented and will evolve to meet business
> >needs
> > >>of the users (where if I am a design engineer, my business need is
>to
> > >>find the technically correct solution engine).
> > >>
> > >>This represents a much ignored need as well as a terrible run-on
> > >>sentence. :-)
> > >>
> > >>Ken
> > >>
> > >>At 04:10 PM 7/11/2005, Frank McCabe wrote:
> > >>>Ken:
> > >>>
> > >>>   I think we are in violent agreement. The essence of the Java
> > >>>execution engine example is that it is precisely that the
> > >>>implementation is important -- not whether or not its good style!!
> > >>>
> > >>>   Incidentally, if it is important to know who wrote the
> > >>>implementation, then that too should be part of the public
> >description.
> > >>>
> > >>>   Perhaps another aspect concerns the on-the-wire vs. resource
>model
> > >>>of service semantics. In the resource pov, issues of implementation
> > >>>(transparent or otherwise) figure prominently. In the on-the-wire
> > >>>pov, you only look at the message traffic; and do not even consider
> > >>>the implementation.
> > >>>
> > >>>  They are related of course: a message is valid if it is well
>formed
> > >>>(in some communication language) and if the implied predicate is
> > >>>satisfied (there really is money in your bank account).
> > >>>
> > >>>  But, a Service Oriented Semantics (SOS) would focus on the
>on-the-
> > >>> wire interpretation and require that the service participants
> > >>>faithfully reflect the messages in their systems. I.e., it is
>similar
> > >>>to the model/proof theory distinction: provided that the service
> > >>>implementation is a correct Interpretation of the communication,
>the
> > >>>participants can focus on the Formulae being communicated.
> > >>>
> > >>>That all being said, there are definite limits to descriptions.
>I.e.,
> > >>>there will always be unstated (unstateable?) assumptions that are
> > >>>required to be shared for successful interactions.
> > >>>
> > >>>Frank
> > >>>
> > >>>On Jul 11, 2005, at 12:49 PM, Ken Laskey wrote:
> > >>>
> > >>>>Frank,
> > >>>>
> > >>>>Opacity/transparency is not a dimension on which to distinguish
> > >>>>services;  it is more of a fallacy that we may need to explain our
> > >>>>way around.  This ties in very much with what I've tried to say
> > >>>>about metadata and how metadata characterizing an instance is
>often
> > >>>>the basis upon which the instance is evaluated in the context of
> > >>>>assumptions, constraints, and policies (a/c/p) of providers and
> > >>>>consumers.  In some circumstances, the amount of information that
> > >>>>needs to be published about an entity may maintain little in the
> > >>>>way of its opacity, but that is more likely related to a/c/p than
> > >>>>it is to WSDL.
> > >>>>
> > >>>>See more inline.
> > >>>>
> > >>>>At 03:23 PM 7/11/2005, Frank McCabe wrote:
> > >>>>
> > >>>>>I read the revised section on services. And an issue that made me
> > >>>>>uneasy before has resurfaced -- about opacity.
> > >>>>>
> > >>>>>I think that the opacity/transparency dimension is not a good
>basis
> > >>>>>for distinguishing services. I am aware of the intuition: that
> > >>>>>somehow when you use a service you should not have to worry about
> >the
> > >>>>>implementation. But what does it mean to say: "the implementation
> >is
> > >>>>>hidden from the service consumer"? If the service is to add lists
> >of
> > >>>>>numbers then its implementation is hidden provided numbers are
> >added
> > >>>>>up correctly. (I.e., we should not care about whether its in C++
>or
> > >>>>>C#; multi-threaded or federated).
> > >>>>
> > >>>>So here the service adds numbers and while we can imagine what is
> > >>>>happening, we care no more of the specific than we do when we
>punch
> > >>>>numbers into a calculator.
> > >>>>
> > >>>>
> > >>>>>But, some services are explicitly about implementation issues:
> > >>>>>security services, workload distribution services, java remote
> > >>>>>execution engine services etc. etc. I.e., it is of the essence
>that
> > >>>>>to use a Java remote execution service, you have to send it a
>java
> > >>>>>class to execute. In that sense, the implementation is critical.
> > >>>>
> > >>>>But you call a java remote execution engine when you have an
> > >>>>appropriate java class to execute.  If it is a general service
>that
> > >>>>is implemented as a java remote execution then the WS interface
> > >>>>should more likely be a collector of information that (inside the
> > >>>>service) is rewritten as the necessary java class.  You could
> > >>>>require a java class as input but I'm not sure that would be
> > >>>>considered "good style".
> > >>>>
> > >>>>
> > >>>>>A clearer distinction is public/private. A service is
>characterized
> > >>>>>by a public description of its functionality. The public
> >description
> > >>>>>should be sufficient to be able to successfully interact with the
> > >>>>>service; and no additional information should be required.
> > >>>>
> > >>>>Not really because what you consider part of the private
> > >>>>information I might consider very pertinent to deciding whether
> > >>>>your service is suitable for my purposes, i.e. I need it to be
> > >>>>public.  For an extreme example, I used to work for your company
> > >>>>and I know one of your developers is clearly incompetent and I'd
> > >>>>never use anything that idiot touched.
> > >>>>
> > >>>>The challenge is in defining "should be sufficient to be able to
> > >>>>successfully interact with the service".  I think we need to
> > >>>>discuss opacity in these terms because it is strongly connected
> > >>>>with SOA and may be more useful as a vague concept than as an
> > >>>>absolute measure.
> > >>>>
> > >>>>
> > >>>>>Frank
> > >>>>
> > >>>>Ken
> > >>>>
> > >>>>
> > >>>>
> > >>>>
> > >>>>
> > >>>>
> > >>>>--
> > >>>>
> >
> >>>>--------------------------------------------------------------------
> >--
> > >>>>-----------
> > >>>>   /   Ken
> > >>>>Laskey
> > >>>>\
> > >>>>  |    MITRE Corporation, M/S H305    phone:  703-983-7934   |
> > >>>>  |    7515 Colshire Drive                    fax:
> > >>>>703-983-1379   |
> > >>>>   \   McLean VA
> > >>>>22102-7508                                              /
> > >>>>
> >
> >>>>--------------------------------------------------------------------
> >--
> > >>>>------------
> > >>
> > >>--
> >
> >>----------------------------------------------------------------------
> >-----------
> > >>   /   Ken
> > >> Laskey
> >\
> > >>  |    MITRE Corporation, M/S H305    phone:  703-983-7934   |
> > >>  |    7515 Colshire Drive                    fax:      703-983-1379
> >|
> > >>   \   McLean VA 22102-7508
> >/
> >
> >>----------------------------------------------------------------------
> >------------
> > >>
> > >>
> > >
> > >
> >
> >--
> >
> >-----------------------------------------------------------------------
>-
> >---------
> >    /   Ken
> >Laskey                                                                \
> >   |    MITRE Corporation, M/S H305    phone:  703-983-7934   |
> >   |    7515 Colshire Drive                    fax:      703-983-1379
>|
> >    \   McLean VA 22102-7508
> >/
> >
> >-----------------------------------------------------------------------
>-
> >----------

--
      ---------------------------------------------------------------------------------
   /   Ken 
Laskey                                                                \
  |    MITRE Corporation, M/S H305    phone:  703-983-7934   |
  |    7515 Colshire Drive                    fax:      703-983-1379   |
   \   McLean VA 22102-7508                                              /
     ----------------------------------------------------------------------------------