SOA (Service Oriented Architecture) once seemed to promise a new world of smooth cross-platform and cross-language interopability, high software reuse, easier maintenance of complex systems, and clean wrapping of legacy systems. Has it delivered? I found the recent Microsoft Architecture Insight Conference surprisingly downbeat. Sam Lowe from Capgemini gave a brisk overview of where SOA is valuable, emphasizing that it is not always applicable, that its value is hard to prove, and that it often does not live up to its hype. “You need to think across business and IT”, he said, making the point that project roadmaps should not be technology-centric. It is no good rewarding people simply for creating services; if you do, you end up with lots of services for no clear reason. Too many services may be worse than too few.

All sound stuff. A second session from Conchango’s James Saull continued the theme in his “real world” session. It’s “very very rare” to see SOA success stories, he told us, following up with “I have never seen a business case for doing a service-oriented engagement.” One delegate immediately came up with one; but there was a fair amount of head-nodding as well. The supposed reusability of SOA services also got bashed. “I haven’t seen anyone to date really getting reuse,” said another delegate. Versioning and dependency issues were mentioned. The takeaway was not that SOA is useless, but rather that resources have been wasted in a mistaken belief in SOA as a solution to everthing.

It took a case study to bring relief from these depressing assessments. This was from the London Underground, the same WPF (Windows Presentation Foundation) application that I commented on earlier, but with a little more detail. I was not the only person impressed by this application; apparently the governor of the Bank of England, Mervyn King, paid the developers a visit to find out more about it. Although the project is only a proof of concept, there is great enthusiasm for it and a production version is actively being planned, though it will take until Q3 2007 before the 20,000 London Underground desktops are powerful enough to run it (.NET Framework 3.0 is required). Passengers may actually see station displays running WPF.

The London Underground application brings 3 or 4 systems together into one visualization. You can think of it as a kind of Enterprise mash-up. Although it is the user interface that catches the eye, it would not be possible without an existing investment in SOA, going back several years. It therefore appears that London Underground is getting reusability and other SOA benefits that are eluding most others. I asked what the secret was. The answer was a little vague. “We’re fortunate that we had the right services in the right place at the right time,” said developer Keith Walker. Peter Goss expanded on this. “We have four of five main services we use, but each of our large applications has an interface exposed which we can consume from if necessary. It’s an ongoing process.” In other words, every application was designed to be part of a platform, not just to work in isolation.  There was the right level of granularity for the services, which matched the business well.

Here at last is an example of SOA yielding perhaps unexpected benefits, presuming that the proof of concept does translate successfully into a deployed application. For more background on this case study, download the presentation referenced in my earlier post.

So what does it take to be successful with SOA? It’s hard to discern whether London Underground is just a particularly good fit for this kind of architecture, or whether it happens to be using development principles that would work equally well in other contexts.