12 September 2008
Can SOA be done with an agile approach?
I don't delve too much into the cluttered world of SOA (ServiceOrientedAmbiguity), but I get this question often enough (in some form or other) to be worth a pontification.
When I first came across agile software development (in the form of Extreme Programming) the most troubling aspect for me was its approach to software design. Like many I'd become used to the notion that software should be designed before it was programmed, while XP seemed to encourage an almost wilful embracing of design ignorance. In 2000 I was asked to give the closing keynote at the first Agile/XP conference, and in order to gather my thoughts I ended up writing Is Design Dead? - an essay that examined the fate of design in an agile world.
I'm still quite proud of that essay and I think it's well worth reading today - but for this bliki entry I'll summarize. I talked about two driving approaches to software design: planned and evolutionary. Planned design works out the design of software in one phase and builds (programs) it afterwards. In this case changing the design is hard once you've begun construction. Evolutionary design assumes regular change of the design even once you've done significant programming. I concluded that the practices of XP provided a disciplined approach to evolutionary design, thus making it much more practical than people had realized. This change did not remove software design (it is not dead) but did significantly change how we think about design.
The argument for planned design in an SOA context is that we are building webs of interconnected, loosely coupled systems. In this situation each system is making its services available as a PublishedInterface to the whole enterprise. Published interfaces are hard to change, therefore you need planned design to get them right so you don't have to change them. Planned design is also a reaction to the chaotic system interconnections that people see in most organizations. This chaos is the result of poor design, so the feeling is that better planned design will prevent this happening in future.
Evolutionary design is an essential aspect of agile methods. One of the key foundations of agile methods is the desire to handle, indeed welcome, change. Planned design assumes change is hard, and thus tries to predict where it occurs. If changes occur within the predicted boundaries then it's easy, but if it falls outside those boundaries you're out of luck. Agile thinking assumes unpredictable change is inevitable and tries to enable it in a controlled way.
So as I look at SOA, or any other design context, the fundamental question I ask is "is change predictable?" Only if change is predictable is a planned design approach valid. My sense is that if predictability is hard within the context of a single application, it's doubly hard across an enterprise. If we use planned design in a unpredictable context we find that however good the plans are, they will be undermined by the unpredictable changes, leading to the same mess we are currently in. Usually, however, the mess is worse because a there is significant sunk cost in a planned design, this can easily reduce the business value that an SOA effort is intended to produce, particularly if time-to-market is important.
As a result I think we have to bite the bullet and figure out how to do evolutionary design in this loosely connected context. After all the whole of point of loose coupling is to make change easier. At the center of this is thinking about contracts in terms of change, with such ideas as Consumer Driven Contracts.
This direction leads to things like Jim Webber's notion of Guerilla SOA. This builds up SOA using small steps directed by producing business value. Since producing business value is the whole point, this offers a path to producing a much better return on investment. It's certainly an approach our clients appreciate.
Can evolutionary design scale to SOA sizes? In my view we have an existence proof at a much larger scale than even a big SOA effort - the web itself. The web is built on very loose coupling and lots of unpredictable changes. It is, in many ways, a mess - but it's a mess that works, delivering real value to lots of people every day.