These are the old pages from the weblog as they were published at Cornell. Visit www.allthingsdistributed.com for up-to-date entries.

March 11, 2004

The no-feedback WS-TX workshop

Yesterday was the feedback workshop on the ws-coordination, ws-atomictransaction and ws-businessactivity specifications. There were about 40 people in the room from big and small players; MS, IBM and BEA of course as the spec authors, but also people from Iona and Arjuna, who have specs in the WS-CAF framework, and also the folks from Choreology who were the main editord of the OASIS BTP spec.

I expected that such a mixed crowd would give quite a bit of fireworks, but I couldn't have been more wrong. Hardly any disagreement about the hard stuff in specs, and the items that ended up on the discussion board where mostly clarifications and points where the some ambiguity could arise form the current spec, and some discussion about fault codes and the state transitions. Already quite quickly the meeting's non-buzz fell under my attention span threshold. The only people trying to push harder were the guys from Choreology who had requests for particular extensions in BA that they were advocating. But I had this feeling that the current scope of the specs was already cast in stone and that the workshop would not do much to change any of the fundamentals of the specs.

Eric Newcomer made a plea for collaboration between the CAF and TX people and there were friendly nods around the room but no real reaction. It felt like the geriatric department of the local hospital where everyone, including me, was dozing off after their medications. The meeting even finished early, which was a big contrast with the ws-eventing feedback meeting were the discussion could have gone on into the night if the organizers hadn't kicked us out.

I can hardly imagine that the spec authors were much happier with the limited feedback, but I could be wrong of course.

Posted by Werner Vogels at March 11, 2004 03:28 PM
TrackBacks

TX Feedback Workshop
Excerpt: We are just working our way through the post-workshop follow up actions after the Web Services transaction workshop last week on the WS-Coordination, WS-AtomicTransaction and WS-BusinessActivity specifications. As organizers, we rescheduled from the o...
Weblog: TheArchitect.co.uk - Jorgen Thelin's weblog
Tracked: March 16, 2004 02:01 AM
Interesting comments from Choreology's CTO
Excerpt: Very interesting comments from Choreology's CTO at Werner Vogels's blog....
Weblog: Stefan Tilkov's Random Stuff
Tracked: March 18, 2004 05:55 AM
Comments

Rashomon is a good movie.

I think your attention may have lapsed to such an extent that you missed some of what was going on.

Choreology had to face down an aggressive attempt by the moderator of the feedback session to prevent us speaking to give Choreology's synthetic feedback (which is based on implementing BTP and WS-C+T, atomic and BA varieties).

Moderator: "I'm not going to let you get up here and speak if you want to turn WS-BA into BTP". Surprisingly defensive for the representative of one of the largest computer corporations. In the end we were offered 10 minutes speaking time, which might account for not getting to the bottom of the rather large list of items we wanted to raise. However we were invited to post more detail in written form, and I think the resulting discussion will be interesting. I don't think the issues surrounding BTM are at all well understood, and it will take some time to tease out the discussion you're looking for.

In this respect, this workshop (and parallel activities such as BPEL and the Grid research group on transactions) are dealing with something more substantial than respraying CORBA/COM/MOM with SOAP and WSDL, which I personally find the closest thing to Mogadon.

The issues that were raised go far beyond bug-fixing.

Example: are business promises to be protocol promises? In other words, does a reservation of business resources amount to a recoverable, callable promise reflected in the BA protocol? Or is it an ephemeral promise that precedes use of the BA protocol (and putting into question the utility of that protocol).

Example: should all entities that promise resources (offer a quote, make a bid etc) be confirmed or cancelled en bloc (uniformly)? Or should a consuming application be able to cancel some and confirm others? This latter ability (what BTP calls a cohesive business transaction, or "cohesion") is vital for modelling real-world BTs, where some elements are critical, and some are optional, and where there are therefore several viable, "successful" outcomes.

Example: does business transaction management have application outside the super-constrained case of BPEL-based compensations, which was presented as the sole motivation for WS-BA. E.g., on the Grid? In non-workflow coordinated activities? To create composable application services?

It was also interesting to see the lack of what the execrable Tony Blair calls "joined-up thinking" over WS-BPEL and WS-BA. Informed observers of the WS-BPEL standards committee in OASIS will know that senior technical representatives of IBM and Microsoft have cast doubt on the need for any kind of separate transactional protocol in that forum. Whereas the same companies' equally senior technical representatives in this feedback workshop are (rightly) concerned that application or process designers are not burdened with rewriting BTM software in BPEL - a challenge for even the most agile of hackers.

Of course, if you are right that WS-AT and WS-BA are cast in stone (which was not my impression from the meeting), then an opportunity to bring about convergence of BTM standards will be lost.

The current proliferation (BTP, CAF, WS-C+T) is an impediment in developing application-level coordination and process synchronization solutions, and it's clear to us from our customer contact that this is a sore need.

At the meeting I said that BTP is dead as a contender for leading the way in Web Service transaction management. Let's hope that rigidity or lack of willingness to make minor textual changes with sometimes far-reaching methodological implications doesn't vitiate our goal of making WS-BA the obvious way of underpinning transactional business processes and composable apps.

(A further benefit of WS-BA leading the way -- with useful additions beyond the current draft -- would be to prune off the highly repetitive parts of WS-CAF, e.g. ACID and LRA, which add little if anything to that already achieved by BTP and WS-BA.)

Alastair Green, Choreology CTO

Posted by: Alastair Green on March 15, 2004 10:29 AM