Thursday, 16 April 2009

Fartlek development

The amusingly named Fartlek training is a form of athletic training in which high intensity work and low intensity work are alternated in the course of each training session. The two forms of work are complimentary in that they stress the you in different ways (anaerobic and aerobically) and allow you a chance to recover.

I think that this is a good model for software development in software start-ups.

Priority one: Build something that people want, fast

The main risk in a software start-up is making something that people don't want. So you want to find out whether your initial idea has merit, fast. So -- if at all possible -- you build a small prototype, and test it out in the marketplace, fast.

In order to satisfy the need for speed, you cut a few corners, and because it's small (and your team is good) you can get away with this approach (for now).

If no-one wants your prototype, you scrap it, and you haven't wasted time on lots of process that was supposed to pay off in the long-run (because there is no long-run).

On the other hand, if it the prototype is promising, you will learn from the market feedback how to make it better, and it will start to grow. This is where it is critical to switch gears and slow down (at least for a while). Because if you do not, you will find yourself in the realm of cowboy development, the bugs will start to bite, and the pace of development will drop as your team gradually spends more and more time debugging, less time developing, resulting in a drop of confidence -- "I don't want to quickly make this change because it might break something" -- and loss of tempo.

Priority two: Restore quality (and sanity)

The alternative is to switch modes. Fred Brooks said "build one to throw away", but at a minimum you should review what has been done to date and investigate and prioritize the very real risks of continuing with a prototype. These might well include:
  1. Customer or prospective customer loses data or has it corrupted
  2. Customer or prospective customer experiences bugs, crashes, or loss of service
  3. A crash or a bug during a demo
  4. The software does not scale, so performance degrades over time
  5. Development slows so that critical new features can not be delivered in a timely fashion
Improving the quality of the software process is what is needed for most of these, and many of the kinds of measures that help have largely been formalized in the Agile development processes:
  • Pair programming: Two heads are better than one, plus it offers real time code reviews, mentoring, camaraderie and friendly competition, less web-surfing
  • Test-driven development: Writing automated tests clarifies the design and as they build up offer an early warning system when a change breaks something unexpectedly (Design by Contract gives similar benefits)
  • Continuous integration: Frequently "checking-in" and automatically running the automated test-suite identifies incompatible changes made by different programmers or pairs sooner rather than later; saves you from merge hell
  • Refactoring: By reviewing code, eliminating duplication, reducing coupling, and improving the use of abstraction, your software base will become more concise, readable and maintainable
  • Short iterations / planning game: By maintaining a list of feature requests and bug-fixes that the developers estimate and the internal customer prioritizes (within a budget based on the amount of work accomplished in the previous iteration), a sustainable rhythm is established, and healthy prioritization is forced
  • Sustainable pace: Working more than an eight hour day / forty-hour week leads rapidly to damaging the code base, loss of morale, and bull sessions as people are present in body, but not in spirit -- avoid the "Death march" scenario
Additionally, a design review will be needed to knock the existing architecture and code into shape, and tests will needed to be added retrospectively if you didn't practice test-driven development in the initial prototype phase.

For a web-based service additional infrastructural elements and diagnostics should be reviewed and added.

All of this is manageable as long as the prototype is small, but the longer these steps are delayed, the greater the costs that must be paid. It's a lot like going into debt. You not only have to repay the loan, but also the accumulated interest. Some people call this design debt. (And the interest rates on design debt tend to be high.)

And then the cycle turns

While making incremental changes continued agile iterations is probably the way to go, slow and steady (but not slowing down). When the time comes to make a leap forward, knocking up a quick prototype is a great way to vary the routine, and makes business sense.

Almost everyone who learns to program first experiences the joys of "cowboy development" on small personal projects, but when they "turn pro" or start working on larger projects they soon learn that it doesn't scale beyond what you can easily hold in your head.

Some people think that up-front design and analysis will save the day, but this leads to waterfall with its 1/3 design, 1/3 development, 1/3 testing rule-of-thumb when done well. But then you discover that the requirements were wrong (they always are), and another long iteration is required.

Smart people rediscover short iterations, and lately these have been articulated as Agile Methods, and also include other ways to make the computer (e.g. automated tests) and human nature (e.g. pair programming) work effectively and more satisfactorily. The cost is a bit of deferred gratification by having to do some things up front (e.g. writing tests, refactoring) that pay off over time.

However, it would be wrong to say that an Agile approach is the best for everything, at every scale. In particular, in knocking up a quick prototype one may well be justified in reverting to some of the cowboy practices for a short period. Note that it is not compulsory to drop all the agile practices when doing so -- do what suits the situation.

The tricky bits are:
  1. Not leaving it too long
  2. Paying off the accumulated technical debt
  3. Getting back into the discipline of following the practices
  4. Combating the likely organizational pressure to continue at the previous pace!
So good luck with the fartlek development*, and let me know what you think.

*My esteemed colleague -- and former Olympian -- Ben Loft points out that interval training may offer a better analogy, but the name isn't as memorable.

No comments: