Engineering Architecture: A Syllabus?

Assembling a reading list on the theory of engineering architecture.

After spending a week trying to figure out what to say in one lecture, I realized I could teach an entire class on architecture. In fact, in hindsight, that’s what this class should have been! Oh well. I knew this from the outset, but I couldn’t figure out how to stitch a syllabus together last December. Inevitably, I had to work my way through a full semester of cleaning out the skeletons in my learning-for-control closet before I could figure out what I wanted to dive more into. To be clear, this is a success story and a model of how teaching classes ought to go.

Given that I’m already committed to a schedule for next year,¹ a class on architecture will have to wait a bit. But nothing stops me from putting together a syllabus now, right? While preparing this week’s lecture, I assembled an unfortunately long reading list for this hypothetical class. Synthesizing this material promises a very interesting story. Let me explain how I’m thinking about its contents.

Though architectural theory is figured out on the fly, adapting existing systems to manage newfound complexity, there are repeated patterns that we can extract from our contemporary human-cyber-physical infrastructure. The architecture class would attempt to synthesize the design principles needed for enabling diversity and error handling. The paper I referenced yesterday by Matni, Ames, and Doyle takes a stab at this sort of view, but I want to look beyond robotics. I’d want to cover as diverse a set of applications as possible while still maintaining some degree of cohesion.

I’d probably start with computing systems. You’ll get different answers about what is needed to build good architectures in your operating systems, programming languages, and networking classes. And maybe you should. I’ll keep repeating myself: I’m not convinced that there’s a “universal theory” of architecture. However, that doesn’t mean we can’t move up a layer of abstraction and draw the common threads together. What are the shared patterns in hardware, software, and network design? I’m particularly interested in studying the 75-year development of software from manually mapping bits on registers to the complex high-level languages of today. There are a lot of interesting theories on modularity, abstraction boundaries, and protocol design, and those should be thrown into the mix.

I’d also extend downward into the physical layer, adding a “cyberphysical” systems view that connects to larger systems like the power grid or transportation network (good references on these two topics are currently missing from the bibliography). I’d spend time on the history of architectures in robotics and control, where we have settled on a platform, arguably by the discipline-wide adoption of the Robot Operating System. The principles were there in the Apollo project: a separation between low-level control, sensing, navigation, and mid-level feedback, and high-level planning. There have been other proposed architectures, like Brooks’ subsumption architecture, that didn’t gain much traction beyond the Roomba. There is something inescapable about the standard three-level architecture, and I want to unpack more about this diversity-enabled sweet spot.

I would like to examine some architectural theories in systems biology, especially those of Gerhart and Kirschner. We’d have to at least read some parts of The Plausibility of Life, mostly because it’s really good. I also think that we do learn a lot by reflecting our technology onto biological systems. I’m sure we’ll find interesting examples and insights by seeing how others have done this.

I also want to look at how we engineer architectures for organizing people. The complexity of the corporation and the computer grew symbiotically, and there are clear influences of human organizational behavior on information technology. There’s clearly a co-evolution of computing architectures with human architectures. Herb Simon, who has had as much influence on management science as computer science, would be a key figure here.

And since I can’t pass up an opportunity to dig into more Cold War technological history, we’d look at some classic theorizing about complex systems. This class would not be an aughties-era complex systems class. But I’d like to find the point in time before the network science people split off from the cyberneticists. So we’d go back and read Wiener and Weaver, Ashby and Simon, and look at what they got right and what they missed.

The bibliography needs both growth and pruning. But I have plenty of time to get it in order. Help me flesh it out. What topics and references would you add? What other books on architecture, organization, protocols, and design should I throw on there? I’d love to see your suggestions in the comments.

1. I’ll teach “Forecasting: WTF?” in the Fall and probability in the Spring.