Hi Cory, (01)
After spending considerably more time on this today than I planned (when
it wasn't interrupted by "real" work and meetings), I honestly think
we're trying to accomplish the same goals. However, I think there are
currently two things which are possibly preventing clearer
communication. (02)
First, I think we have a bit of a terminology impedance which is
actually caused by the second: different starting points to SOA which
may have resulted in different perspectives. I believe that we can get
around these issues. What I will do is walk through some of the points
from the conversations over the last few days and then see if I managed
to catch the correct babel fish. However, there are still some things I
don't understand. Please forgive me if I'm just being thick. (03)
Based on your past published work [1,2,3], it is quite clear that you've
put a lot of thinking into where you are now. It's also pretty clear
that you've an extensive background with CORBA, UML and MDA. (04)
>From what you've said, I think that this has had a distinct impact on
how you view SOA. As you also said, you have seen and been a part of
several SOA solutions implemented in CORBA. (05)
If I understand what you've been saying over the last few emails, I
understand you to have the following definitions. I'm trying to nail
these down so that I'm sure what you mean, and I don't make the wrong
assumptions. I realize you may think I'm silly, but I really think it's
important to get straight. (06)
community -- a pair of roles interacting for some purpose (07)
role -- a task or responsibility within a community (08)
interaction -- the means of communication between a role (09)
community contract -- the constraints and manner of interaction between
a pair of roles using specified service interfaces (010)
specification -- codification of a given community contract (011)
SOA -- an architecture consisting of the following core elements:
- people and organizations
- roles and responsibilities
- interactions between roles
- community contract (012)
service interface -- a set of methods exposed by a distributed object
(as exemplified by middleware technologies like RMI/IIOP, CORBA, DCOM,
.NET remoting and SOAP) (013)
collaboration -- interactions between roles to realize a business
process (014)
integration -- the combination of all of the elements defined by the SOA
(above) to enable a collaboration (015)
interoperability -- the level to which a role and service interface
conform to the interaction contract of the community within a
collaboration (implying some sort of shared context between the two
roles) (016)
There are also a few terms that I'm not sure I understand from your
usage. (017)
agility -- this is a term which has a number of different meanings,
depending on your context. Can you explain to me what you mean by: "how
things work together for a common purpose while retaining _agility_"? (018)
architecture -- I think it's the environment in which all of the above
interactions take place, but I'm not really 100% sure from your usage. (019)
If the above are right, or even close, I am now starting to see why
we've not been effectively communicating. :( (020)
I have done a few distributed computing projects as well, including some
with J2EE, CORBA, DCOM, .NET remoting and even some simple PDO things
under NEXTSTEP, XML-RPC and basic SOAP/1.1. So I am pretty well versed
with the paradigm. (021)
However, for the last 18 months, I've been implementing part of an
on-going SOA project based nearly exclusively on asynchronous
messaging. Based on this experience, and from reading everything I
could find about Web services, loose coupling, messaging, ESB and SOA,
I've come to the distinct conclusion: (022)
SOA is not about distributed objects. (023)
This is a pretty fundamental notion on which we seem to differ, but I'll
get to this in more detail as I go. (024)
Even though I'm not in 100% agreement with all it says, the W3C Web
Services Architecture [4], along with the W3C Web Services Glossary [5],
provide some useful, public definitions of terms. Like design patterns
enable communication of architectural design concepts between software
architects and developers, it is useful to have a baseline vocabulary
when discussing these things. And for everyone in the wings with a
semantics & ontology background, in case you were wondering, I haven't
missed the irony of that last sentence. :) I have tried to make the
terminology I use consistent with these definitions. (025)
I'll not go into all of them, but, to me, SOA is about "business"
services, but, like most things, what a "business" service actually is
depends on your point of view. If you're part of the process that
implements or is responsible for the desired outcome, you probably see
all of the little steps required to carry it out. If you are not part
of the process, you see the entire process as the service. This is just
natural human abstraction when they don't know or don't really care
about the details. (026)
For example, you (the requester) are hungry. Not knowing any better,
you pull through the drive-up at McDonald's (the provider) and order a
value meal, pay your money (send a complex message), pull up to the next
window and wait. Eventually, someone appears at the window with your
cholesterol bomb (receipt of a complex message), and you drive away
happy. (027)
This example seems to map to your definitions as the following: (028)
you (the requestor) -> role #1
McD's (the provider) -> role #2
you + McD's -> the community
hanging out your car window and talking -> the interaction
giving your order (including the money) + receipt of food -> community
contract
McD's window + your order + money + bag o food -> service interface
you ordering food from McD's -> collaboration
all of the above -> SOA (029)
Using this simple scenario again, here's how I would define the various
bits (more-or-less from the WSA/WSG): (030)
you -> the requestor
McD's -> the provider
ordering food -> the service
hanging out your car window and talking -> the transport
your order + money -> request
food -> response
send (any order for food + money); receive (food) -> service interface
roads + traffic laws + houses + shops + fast food places -> SOA (031)
Just to clarify some of these key differences a bit more. The way I see
SOA is the environment in which a service exists. This service can be
big or small, but it provides the same sort of granularity, regardless
of the business process being achieved--it is a business service. (032)
Also, services can be complex or simple. If we were in the pub, and I
asked you to go get me a burger, I'd give you my order plus some money
and get a burger back from you. I wouldn't care if you needed to go to
three different places to find what I wanted (except that it might be
cold). (033)
In this case, this would be a complex service (degenerate
case--specifically an intermediary) because you needed to do things on
my behalf. Later, I might also want to go to McD's directly, or I could
ask you again. If life was really good, I could use the same data in
both cases, e.g. both you and McD's would provide the same service
interface--enabling interoperability between service providers. (034)
The SOA part kicks in when I can drive from McD's, to BK, to Wendy's to
Arby's, or wherever, and send the same message and expect the same
response. In W3C-speak, ordering food is the abstract service and each
of the fast food places is a concrete implementation of that service (an
agent). The SOA tells me how I can get from point to point, what rules
I must follow, and maybe even gives me directions. It may also allow me
to get fuel, buy a candy bar or even a newspaper--if those messages
existed and there was an agent implementing a service providing them. (035)
The way I understand Web services and SOA is that they provide these
higher-level interfaces to business services, the invocation mechanism
and the messages exchanged. You get the interoperability because you
*can* place the same order at all those places and expect, with minor
variations, to get the same thing back. If you don't, they don't
implement the same service interface. (036)
I know these things sound very similar to what I suspect you understand,
but to me, they are very different. Messages are not parameters to
remote methods, and WSDL is not IDL. While you can use them that way,
this doesn't buy you anything over any other type of distributed objects
except unmolested access through port 80/443. (037)
The business processes are the interactions between the requester and
provider (through their agents implemented in software), if they're
RPC-style interactions, one-way interactions, or a complex, long running
choreography involving messages going from A -> B -> C -> D -> A as a
directed graph. The business processes can be modeled on any meaningful
level, but the type of centralized architectural view via MDA as I now
understand it (after today's research) isn't really there. (038)
Yes, you have auditing and governance as part of the architecture, but
it's as a peer to peer set of interactions between organizational
entities. Only the provider *needs* to have a service interface,
because the requester can send the request and then ask for the
asynchronous response. They aren't objects. Interoperability at that
point boils down to: (039)
- agreement or following a published business choreography
- agreement on the message data (or any transformations)
- standardized or commoditized transport (040)
If I want to book a flight, I provide the same information over the
phone, on the Web or in person to a travel agent. They all implement
the same service interface, but the transport and agents are different.
To me, this is where you get the power of SOA, and the flexibility to
exchange messages with parties that weren't there yesterday. (041)
I know you all now probably think I'm crazy, but does this make sense?
Is it the same or different than what your understanding is? (042)
ast (043)
[1]
http://www.idealliance.org/papers/xml02/dx_xml02/papers/03-02-04/03-02-04.pdf
[2] http://www.semanticcore.org/requirements/InterfaceAdaptation.pdf
[3]
http://colab.cim3.net/file/work/SICoP/2006-02-09/Presentations/CCasanave02102006
[4] http://www.w3.org/TR/ws-arch/
[5] http://www.w3.org/TR/2004/NOTE-ws-gloss-20040211/ (044)
On Mon, 2006-03-27 at 23:20, Cory Casanave wrote:
> Andrew,
> One think I suspect we can all agree on is that there are multiple
> applications for SOA. Perhaps a little more background on the kinds of
> problems we are trying to solve would be appropriate, below is our focus.
>
> Business Need; Collaboration and integration
>
> The idea of an isolated company, organization, mission or application has
> become the exception. It is simply impractical to think of what you or your
> organization offers without considering the environment in which it is
> offered and performed. Businesses have their supply chains, value chains
> and business processes. Defense has joint missions and collaborative
> forces. Applications require integration and interoperability.
>
> In each case the problem is; how to all these things work together? The
> problems is also that how it works together is a dynamic environment as
> parts and pieces get outsourced, insourced, or modified. The problem is
> also that most solutions to enabling these things to work together get tied
> to particular processes, information or technologies such that it becomes
> brittle and inflexible.
>
> This is the problem we wish to address - how things work together for a
> common purpose while retaining agility. This problem exists at many levels,
> from a worldwide scale where there is collaboration between countries and/or
> major organizations to how people work together within a department. These
> "big" collaborations are realized by smaller and still smaller
> collaborations - until you have a small group executing their specific
> "business process" on their specific information to realize their part in
> the larger picture.
>
> A community may be small or large, the smallest being a pair of roles
> interacting for some purpose, the most common example being a retail
> purchase - buyer and seller. But communities are often large, such as the
> financial management community of a large enterprise.
>
> Where people, organizations, missions or countries work together for a
> common purpose - there must be some element of agreement, some implicit or
> explicit understanding of how they are going to work together for that
> purpose. That agreement may be a simple sales agreement to a complex
> multi-lateral collaboration of many parties. We express this by saying that
> people and organizations are actors playing roles with given
> responsibilities within a community and they enact their common purpose by
> behaving and interacting according to the contract of that community.
> This is the architecture of that community.
>
> We can also look at this community of actors playing roles as each providing
> a service to the community - this is the "larger" view of service, more akin
> to a business service - like maintaining a fleet of vehicles. Thus the
> architecture of the community is the business view of the "Service Oriented
> Architecture".
> Note: This kind of business service of an actor playing a role in a
> community is related to, but not the same as the service interfaces one
> finds in middleware technologies, such as web services.
>
> Between these actors playing roles there are interactions - these
> interactions are frequently bi-directional, asynchronous, choreographed and
> long-lived. They represent the "conversations" between the parties playing
> the roles. Each of these conversations generally correspond a pair of
> service interfaces, one for each side of the conversation. These service
> interfaces can be instrumented in technology, and when they are, you have
> provided a vehicle for that community to operate more efficiently and
> openly. You have provided a way for the "business process" of the community
> to happen without any special monitor or control, but as the natural
> progression of actors playing their role based on the community process.
>
> So for us, from the perspective of a "community" the interesting
> architecture is how all of these actors work together, and by extension, how
> all the services work together for a business purpose. A single service,
> out of context, can provide value - but an architecture of a community of
> roles and services can transform an enterprise or influence a society.
>
> It is this kind of community we would like to demonstrate - how it can be
> described, architected and realized with SOA. The paradigm we use to do
> this brings together standards based collaboration modeling (Based on
> OMG-EDOC), information modeling (Based on UML) and process modeling into a
> cohesive architecture for a community.
>
> A specific example is an architecture we have just done for the GSA'a CFO
> for their Financial Management Line of Business - how can GSA organize to
> provide financial management services within GSA and externally. This is
> what we would consider an SAO in that it describes these communities, the
> roles and interactions, the information models and processes. (You may also
> consider that this is more than an SOA in that it also shows how some of the
> roles are implemented, but at the top is the SOA). This SOA does not
> mention any systems, middleware or technology - it is how the business is
> understood as collaborating business services. We then derive (Using MDA),
> how middleware services interfaces can help realize and implement these
> communities. There are about 60 roles and hundreds of service interfaces in
> this architecture (not monolithically, they can be understood in smaller
> pieces).
>
> So, this kind of architecture for a community where the goal is to
> understand how roles interact using service interfaces to achieve a common
> purpose is a great way to utilize SOA. It allows different actors,
> technologies, systems and processes to be used "under the covers" to help
> the actors play their role. This kind of community simply does not function
> without some agreement of this kind, without a community contract. You
> can't do "financial management" one WSDL interface at a time.
>
> This kind of community architecture can be differentiated from applications
> where a single "web service" can supply a unique capability without this
> kind of context. These, of course, exist and can provide value - it is
> another application of SOA. But the architecture here is focused on the
> individual service interface, less so on the environment or community.
>
> The demo outline we presented is a minimalist and simple community, but
> intended to show this community effect based on the business need for
> collaboration and integration. You could, of course, come up with any
> number of communities to demo. Another kind of demo could be based on other
> business needs or approaches to applying MDA - I think some of the other
> ideas are more along this line. Our business stakeholders should help guide
> us on what kind of problem will resonate with the business community.
>
> Regards,
> Cory Casanave
>
> > -----Original Message-----
> > From: soa-forum-bounces@xxxxxxxxxxxxxx [mailto:soa-forum-
> > bounces@xxxxxxxxxxxxxx] On Behalf Of Andrew S. Townley
> > Sent: Saturday, March 25, 2006 12:03 PM
> > To: Service-Oriented Architecture CoP
> > Subject: Re: SOA Semantic Variation ( was RE: [soa-forum] RE: SOA
> > CommunityDemoCon Call)
> >
> >
> > Hi folks,
> >
> > I've combined some of the points from both Cory and Ken into one
> > response. Please see in-line.
> >
> > On Fri, 2006-03-24 at 22:32, Cory Casanave wrote:
> > > It is this contract of interaction that is the essence of SOA, this is
> > what
> > > allows multiple actors to play roles as providers or users of services.
> > It
> > > is the glue.
> >
> > Absolutely correct, but so is Ken's point here
> >
> > On Sat, 2006-03-25 at 05:02, Ken Laskey wrote:
> >
> > > The essence of SOA is not that we have architected solutions
> >
> > > but that we have the power to architect the hooks for solution modules
> >
> > > that deal with areas which make interoperability difficult now. The
> >
> > > essence of SOA is we can add new services to fill those voids as
> >
> > > technology and knowledge grows, and we can keep multiple solutions
> >
> > > around as long as these are useful. The essence of SOA is that we can
> >
> > > interoperate (to some extent) with people who moments ago we did not
> >
> > > know existed.
> >
> > What Ken pointed out is why the S of SOA is so important. The community
> > is defined by the services that are provided to it. The interaction
> > models of the services themselves define the interaction models of the
> > architecture. This is also where you can get explosive complexity if
> > things within the architecture aren't managed correctly. So Cory's
> > point about the "contract of interaction" is correct, but it exists only
> > between a requester and provider.
> >
> > The architecture defines how these two agents can communicate, with what
> > data models and under what constraints, but it cannot do more than
> > that. Satisfying those constraints and being able to participate in the
> > community require the business relationships that are the glue that Cory
> > mentioned.
> >
> > > How you get to satisfy that contract is another matter. This is almost
> > > always a facade on another system or component or service. How you map
> > from
> > > this "community" specification to your "internal" specification is
> > > interesting and important, but it is not important to the community.
> > This
> > > is your business (So for the demo, this can be a differentiation point
> > for
> > > the implementers). I think some of the conversation has been how to
> > achieve
> > > this façade mapping, which is not then NOT something we have to agree on
> > but
> > > is a great thing to demonstrate.
> >
> > Actually, if this is in relation to my comments, then this comes from
> > the way I expressed it. I was actually not talking about this type of
> > mapping at all. This is what we refer to as "behind the service
> > boundary", and it is, as you said, only really relevant to the service
> > provider and not to the community of requesters. I'll come to more of
> > what I meant below.
> >
> > > The reason we like MDA is it gets us from the business model (providing
> > > context and definition for the processes and messages) to the contract
> > of
> > > interaction (E.G. WS-* and other technologies) with full tracability and
> > a
> > > lot of automation. Just this is a big win - far beyond the current
> > common
> > > practice. Having the business model in the picture tackles many of the
> > > "what does this tag mean" questions without a full ontology.
> >
> > True, but doesn't that pretty-much imply point-to-point interactions and
> > limited re-use of commonalities through standardized message types?
> >
> > These standardized message types may be used by multiple business
> > processes. This way you can attempt to define the message fragment
> > (e.g. a complex element using XML) and state what that message type is
> > intended to be. This says what it is in the context of the entire SOA
> > community. Once you have this done, the service model for each service
> > specifies the context for how it is actually interpreted by that
> > service.
> >
> > The best example of this sort of thing is an address. You can specify
> > the structure of an address and how the elements should be populated (by
> > profiling, rolling your own, or pulling in xAL), but that in one case
> > the address data returned from the yellow pages is for a business and in
> > another case the one returned from the white pages is for a residence is
> > dependent on the service creating the message. It may be represented by
> > the content model as type="business", but ultimately, the context is
> > provided by the service model and the business process it represents.
> >
> > >From what I've seen of MDA from Cory's XML'02 paper "Enterprise
> > Distributed Object Computing" and his referenced presentation, it
> > doesn't really look like MDA can handle this very well. MDA seems to
> > provide higher levels of abstractions to support more generalized
> > process modelling, but it's still dealing with an object-oriented
> > approach once you get down to the information model. Once this happens,
> > you've defined a representation of the concept and you start getting
> > into trouble.
> >
> > I am not convinced that SOA is about objects. Yes, you need a way to
> > ship data around, but the power of shared concepts does not come from a
> > particular representation. It comes from a shared understanding of the
> > concept. If you have this, why not have a shared representation as
> > well?
> >
> > > So the specification we described in the straw man is only the community
> > > contract, with full expectation that interesting approaches will be used
> > to
> > > adapt this community contract to systems and other SOA models. But the
> > > community contract would be our anchor point. The important point here
> > is
> > > that "SOA", as an "architected" solution (the A in SOA) does not need or
> > > expect semantic variation in the contract. But realizing SOA contracts
> > will
> > > be much easier if semantic approaches are available.
> >
> > The other thing is: once you have a community, you must have some sort
> > of governance of that community (even if it's 'none'). This governance
> > model decides how and when shared data models can change. It isn't
> > about semantic adaptation at the service boundary, it's about changing
> > or extending a data model used by many services in the community. You
> > can't cause everyone to change their agent implementations because
> > someone has a legitimate need for including more information into a
> > common model. I believe that this issue is what both Ken and I were
> > trying to highlight (Ken, please feel free to correct me if I'm wrong
> > here).
> >
> > What worries me about what I've seen and read thus far about semantic
> > negotiation across ontologies is that this is really trying to provide a
> > technical solution to a community problem. It goes back to the
> > governance thing I mentioned above. If you have a community, you must
> > have some shared understandings within that community. Yes, there will
> > always be intersections and touchpoints between communities, but this
> > can be resolved by putting the semantic negotiation at those
> > intersection points.
> >
> > The fact that within a single community you can have a nearly infinite
> > variation in the way the data for an operation is structured and what
> > the operation to use it is called is *not* a good thing. Sure they're
> > all "semantically equivalent" because what they are supposed to do is
> > the same, but how you invoke it should be agreed within the community.
> >
> > Just because I can define them all in WSDL and figure out how to map
> > them does not make it legitimate.
> >
> > Within a community, there has to be some level of shared sense of
> > legitimacy or there is no community. We can't drive on whichever side
> > of the road we want to just because we have a car. The community says
> > that you drive on the right, so you drive on the right, If you could
> > drive from New York to London, you'd need to make some slight
> > modifications when you cross the community boundary, but once you're in
> > the UK, you drive on the left. If you don't agree, e.g. establish
> > legitimacy, there is chaos because everyone's trying to do their own
> > thing.
> >
> > In an SOA, there isn't one contract, there are many contracts; each one
> > defined by the service provider's interaction with a requestor. What
> > tools you use to hook those two together is not a problem for the SOA.
> >
> > The SOA says that you can do it and provides the structure to make it
> > happen. The architecture is relatively straightforward; the services
> > are the hard part.
> >
> > ast
> > --
> >
> > Join me in Dubrovnik, Croatia on May 8-10th when I will be speaking at
> > InfoSeCon 2006. For more information, see www.infosecon.org.
> >
> > **************************************************************************
> > *************************
> > The information in this email is confidential and may be legally
> > privileged. Access to this email by anyone other than the intended
> > addressee is unauthorized. If you are not the intended recipient of this
> > message, any review, disclosure, copying, distribution, retention, or any
> > action taken or omitted to be taken in reliance on it is prohibited and
> > may be unlawful. If you are not the intended recipient, please reply to
> > or forward a copy of this message to the sender and delete the message,
> > any attachments, and any copies thereof from your system.
> > **************************************************************************
> > *************************
> > _________________________________________________________________
> > Subscribe/Unsubscribe/Config: http://colab.cim3.net/mailman/listinfo/soa-
> > forum/
> > Shared Files: http://colab.cim3.net/file/work/soa/
> > Community Portal: http://colab.cim3.net/
> > Community Wiki: http://colab.cim3.net/cgi-bin/wiki.pl?AnnouncementofSOACoP
>
> _________________________________________________________________
> Subscribe/Unsubscribe/Config:
>http://colab.cim3.net/mailman/listinfo/soa-forum/
> Shared Files: http://colab.cim3.net/file/work/soa/
> Community Portal: http://colab.cim3.net/
> Community Wiki: http://colab.cim3.net/cgi-bin/wiki.pl?AnnouncementofSOACoP
--
Join me in Dubrovnik, Croatia on May 8-10th when I will be speaking at
InfoSeCon 2006. For more information, see www.infosecon.org. (045)
***************************************************************************************************
The information in this email is confidential and may be legally privileged.
Access to this email by anyone other than the intended addressee is
unauthorized. If you are not the intended recipient of this message, any
review, disclosure, copying, distribution, retention, or any action taken or
omitted to be taken in reliance on it is prohibited and may be unlawful. If
you are not the intended recipient, please reply to or forward a copy of this
message to the sender and delete the message, any attachments, and any copies
thereof from your system.
***************************************************************************************************
_________________________________________________________________
Subscribe/Unsubscribe/Config: http://colab.cim3.net/mailman/listinfo/soa-forum/
Shared Files: http://colab.cim3.net/file/work/soa/
Community Portal: http://colab.cim3.net/
Community Wiki: http://colab.cim3.net/cgi-bin/wiki.pl?AnnouncementofSOACoP (046)
|