RE: [ws-caf] discussion on approaches for web services/ business transaction

["Haugen Robert" <Robert.Haugen@choreology.com>]

> > > There are good reasons for having specific protocols and
> > > against generic ones
> > Sure. Tightly-coupled internal transactions...
> Now who's being extreme - this is certainly *a* reason
> but is neither necessary nor sufficient.

I'll get a little more extreme below...and even claim that WS-TXM makes
the case for a generic protocol.

> > > and on occassion vice versa.
> > Loosely-coupled external transactions, especially spanning different
> > companies.
> Same as above.

Consider the case of participants (not coordinators) in loosely-coupled
external transactions.

If I have a generic protocol, I can develop generic participants and
[enrol|enlist|register] them in any business transaction.

In other words, those participants would be freely composable.

If I don't have a generic protocol, then I need either different
participants for each protocol or adapters-with-mappings for each
protocol.

The participants would not be freely composable.

> > You could propose a protocol to be included in WS-TXM
> > (even BTP I suppose) that accomplished the things you want to see.

I think even BTP is too complicated for the generic participant.  I
don't see any cost-effective justification for a loosely-coupled
participant to care whether the business transaction is atomic, or
cohesive, or whatever. Coordinator, yes; participant, no. 

I know the argument about wanting more guarantees about participant
behavior, but if it's externally undetectable behavior (e.g. Isolation),
you can't enforce the guarantees anyway.

** Rash Claim **

Actually, WS-TXM has its own attempt at a generic protocol: BP.
Describing interposition between subdomains, the WS-TXM-BP spec says:

"Each domain is represented by a subordinate coordinator that masks the
internal business process infrastructure from its parent. Not only does
the interposed domain require the use of a different context when
communicating with services within the domain (the coordinator endpoint
is different), but each domain may use different protocols to those
outside of the domain: the subordinate coordinator may then act as a
translator from protocols outside the domain to protocols used within
the domain.

"For example, a domain may be implemented entirely using the OASIS BTP
with the interposed coordinator responsible for mapping BP protocol
messages into BTP's atom or cohesion messages and vice versa. Another
domain (possibly in the same overall business process) may use the OMG's
Object Transaction Service (OTS) and therefore provide an interposed
coordinator to translate between the BP model and the OTS. The important
point is that as far as a parent coordinator in the BP hierarchy is
concerned it interacts with participants and as long as those
participants obey the BP protocol, it cannot determine the
implementation."

In which case, assuming the BP protocol could work as described, then it
is a universal protocol.  So either WS-TXM has its own proof of the
feasibility of a universal protocol, or WS-TXM-BP won't work as
advertised.

I think BP is too complicated and contains too many sub-protocols to be
a good candidate for a generic protocol, but that's beyond the arguments
for and against the concept and on to the more-interesting discussion of
what makes a good generic protocol.