November 18, 2009

QCon 2009: Strategic Design

Eric Evans

This is the "least discussed part of DDD, but possibly the most important."

"Not all of a large system will be well designed."

CMW: Multiple people so far have mentioned this: systems exist for reasons. They evolved the way they did for reasons. They were probably good reasons at the time.

Every system has deep assumptions. These are parts of the system that are difficult (or impossible) to change.

Common bad strategies:

  • "Build a platform to make other (lesser) programmers more productive"
  • Cleaning up other peoples' messes, keeping things moving
    • "quality of a design will be a the level of the second-worst developer, since everyone is watching the worst one like a hawk"
  • Rebuilding already working software

Consequences of bad strategies:

  • lead organization into cul-de-sacs
  • obscures feedback to management (enabling bad decisions to continue)
  • make irresponsible hackers into heroes

Strategic design:

  1. Distilling the core domain
  2. Context mapping

    • Core domain: place the most emphasis of work here
    • Generic subdomains (e.g. accounting); you don't want to innovate here (unless you're Enron)
    • Supporting subdomains

Core: what makes your system worth writing? what do you do that distinguishes you?

Contrast Amazon use of ratings vs eBay. They're close in functionality, but Amazon would still be Amazon without them. eBay would not, because they're enabling trust without which nobody would use ebay.

Not core features, but core domain ==> modeling well in our core domain enables us to create great features.

Heroes work in the core domain

  • The people who implement the features that fit on the core domain, from a business sense, might make sense to cleanup after them

CMW: "Strategic" -- identify core, then make it kick ass


  • Example, map of the world from 1600 China. It's at center of the world, everything else around it (but that's not much). Compare with ~1900 NGS (?) map; US is at center; it's using a mercator projection for a reason (plotting a course by ship still works) even though it distorted a number of features.
  • Model: system of abstraction that describes selected aspects of a domain, and can be used to solve problems in it.
  • Serves a particular use -- "cannot stress this enough"
    • NOT as "realistic as possible"
  • There are always multiple models they make different things possible, easy, or impossible

(UML representation of "Blind men and the Elephant")

  • Different models are fine existing by themselves; and then we need to integrate...
    • Models do not have intrinsic meaning, only within a context (Bounded Context!)

CMW: his exlanation of resolving the different elephant models reminds me of Blues Clues. Kids shows get in your head...

Disambiguating models into one is doomed -- these are "Enterprise Models", "one ring to bind them all..."

Resolve model differences thru context map; what are integration points and the nature of the translation between them.

The amount we work on the model depends on where the model sits -- in the core domain?

Precision designs are fragile.

Sophisticated design and modeling require

  • Bounded context
  • Containing a unified model

Anticorruption layer: able to translate between models on different layers.

Do core domain development in a bounded context on a clean platform; you do not need to re-create the world to do this. It may mean you need to create a thick translation layer. This is ok.

Look for assets in the legacy system, think "That's work I don't need to do" rather than "It all sucks"


Next: QCon 2009: Agile Development to Agile Operations
Previous: QCon 2009: Continuous Deployment