Wednesday, October 7, 2009

Beautiful Architecture Chapter 12: When the Bazaar Sets Out to Build Cathedrals

Making the call to completely redesign KDE to fully utilize the power of Qt 4 was, as the author says, a very risky decision. I have worked on a few projects that contemplated such a task, some which made the decision to do a rewrite and some which didn’t.

A project that made the decision to rewrite the system from scratch was a ColdFusion web application that was relatively feature complete. The decision was made to rewrite the system using the most modern .NET framework, ORM tools, and AJAX. The decision makers felt that the project could not successfully continue on as a ColdFusion app due to the relative scarcity and skill set of CF developers compared to those who work in Java or .NET. A major factor in the decision to rewrite the system was the fact that a freeze has been put on existing functionality in the current CF system. This meant that we had a fixed target to shoot at for our base functionality.

A project that did not make the decision was a VB.NET (yuck) web application that had a crippling dependency on hundreds of stored procedures that were hundreds of lines long for all business logic. The complete absence of unit tests made it terrifying to have to implement any change to the system, no matter how small. The system was also constantly in a state of flux, adding dozens of new features every month. The decision was made to make a focused effort to re-architect and upgrade the system in place both because the ideal target framework for their developers was .NET, which entailed few fixes to upgrade from 2.0 to 3.5, and also because the amount of new functionality constantly being added to the system meant that we had a moving target to hit whose base functionality was constantly moving farther and farther away.

I think the bazaar model works for open source projects because the labor is free to the project owners. That means that different implementations of a solution can be implemented and accepted or rejected with no cost to the owners. With an enterprise application where someone is paying the bill for every hour spent on development, a more cathedral-like style of management is necessary up front to ensure that minimal time is wasted on solutions that would not be acceptable in the end project. That isn’t to say that cathedral-style projects never to any prototyping or always get it right on the first try, but the design and architectural guidance provided by the key members should keep waste to a minimum.

That said, I didn’t get the impression from this chapter that KDE doesn’t have strong cathedral influences. The chapter said that high-level design decisions would be made by a core set of decision makers on an infrequent basis and lower-level decisions would be made on a more frequent basis by a more distributed set of decision makers. Although the decision making isn’t entirely centralized, there still sounds like there is quite a bit of cathedral-style planning taking place.

I found that having the chapter focus on both the design process as well as technical implementation details gave quite a bit more context than other chapters, but I also feel that the volumes of information made the chapter feel as if it would never end. I found the ideas of keeping data and control pipelines separate & initially using IMAP to transfer data quite interesting, though I’m sure the former is likely a common practice.

I didn’t get as much from the ThreadWeaver section, nor was it talked about as much during the class discussion. Although it was noticeable shorter than the Akonadi section, it presented an interesting interface for parallelizing multiple instances of tasks while still making it easy for each task to execute sequentially.

No comments:

Post a Comment