Systems Architecture · Chapter 8
Tip
Running example: an air transportation service. Section 8.4 repeats the process for the home data network.
Table 8.1 compiles every question from Chapters 4–7, in the order they’re addressed during synthesis, plus two new ones this chapter answers:
8a. How does the architecture of Level 1 extend to Level 2?
8b. What is a possible modularization of the Level 2 objects?

A simple “no-frills” airline: purchasing → check in → loading → transporting → unloading → check out. Source: Crawley, Cameron & Selva (2016), Fig. 8.1.

Nourishing, entertaining, and crediting (a frequent-flyer program) add secondary value — and make the architecture noticeably more complex, since much of the operational detail must now be replicated for baggage too. Source: Crawley, Cameron & Selva (2016), Fig. 8.2.

The starting point for Level 1 is the Level 0 information: the solution-neutral function, the concept, and the concept of operations. Source: Crawley, Cameron & Selva (2016), Fig. 8.3.
The solution at one level becomes the problem statement at the next.
Tip
Example: “purchasing tickets” (Level 1) becomes the Level 2 intent, specialized down to purchasing online and beyond.

Purchasing specializes into 5 internal processes (linking, learning, reserving, purchasing, amending) and reveals 3 new internal operands and 3 new accompanying instruments — all hidden at Level 1. Source: Crawley, Cameron & Selva (2016), Fig. 8.4.

Applying this zooming to all 9 Level 1 value processes produces 28 internal processes at Level 2 (excerpt shown — ticketing and part of checking-in). Note 3 processes (entertaining, nourishing, crediting) get “demoted” from Level 1 — such judgment calls are routine in architecting. Source: Crawley, Cameron & Selva (2016), Table 8.2.

28 processes mapped onto 22 instruments of form (excerpt shown) — close to one-to-one, with notable exceptions: agents and flight attendants are human, and adapt to many roles. Source: Crawley, Cameron & Selva (2016), Table 8.3.
Three levels are hard to develop and too much to comprehend easily — we usually don’t need to go to Level 3. Examining Level 2 exists to confirm Level 1’s abstractions and inform modularization — not to fully re-architect everything one level deeper.
Tip
Layering lets a network designer view the system through the lens of a single layer at a time — powerful for scalability and robustness, but it makes the functional pathway of the data harder to trace end to end.

The laptop creates an IP packet, encapsulates it into an 802.11 frame with MAC addressing (via ARP), and broadcasts it over WiFi; the WAP receives, de-encapsulates, and re-encapsulates it as an Ethernet frame toward the switch. Source: Crawley, Cameron & Selva (2016), Fig. 8.5.

The switch error-checks, reads the destination, and forwards the frame to the gateway, which de-encapsulates it, performs NAT (network address translation) to swap the private IP for a public one, and re-encapsulates it as PPPoE — modulated by the DSL modem onto the phone line toward the ISP. Source: Crawley, Cameron & Selva (2016), Fig. 8.5.
Three tasks: choose the basis for clustering (processes or form), represent that information (as a DSM), and compute the clusters. Replacing the interaction type (create/affect/instrument) with a simple count of connections lets a clustering algorithm rearrange rows and columns into tightly coupled blocks.
Tip
Here, clustering is based on processes, linked through shared operands: two processes that touch many of the same operands are assumed to be tightly coupled.

The Thebeau algorithm reorders the 28×28 process-interaction matrix so tightly-coupled entities fall into contiguous, shaded blocks along the diagonal — six clusters emerge. Individual cell values aren’t the point here; the block-diagonal pattern is. Source: Crawley, Cameron & Selva (2016), Table 8.4.
Tip
This clustering is organized by operand path, not by chronology — a genuinely different decomposition than the original timeline-based one.
If the airline cared about running time, the time-based organization might make more sense. If it cared about end-to-end reliability per passenger or bag, the operand-based clustering looks more appropriate.
Tip
Reference: Crawley, E., Cameron, B., & Selva, D. (2016). System Architecture: Strategy and Product Development for Complex Systems. Pearson. Chapter 8.
Having analyzed how architecture is built, we turn to real synthesis: defining architectures that don’t yet exist, for complex systems, starting with the role of the architect (Chapter 9).

← Course Home · Systems Architecture · Chapter 8