Network Architecture and Design
The Network Architecture and Design domain tests whether you can pick a topology, an addressing plan and a physical build that meet stated availability, capacity and cost targets - not simply draw boxes on a diagram. Expect constraints (traffic pattern, budget, uptime target, cable distance) and a question asking which design satisfies them.
Choosing a topology
Three-tier (core, distribution, access) fits campuses where north–south traffic dominates and the site is large; a collapsed core merges core and distribution for smaller sites. Spine–leaf (a two-stage Clos fabric) is chosen when east–west traffic dominates, as in data centres: every leaf connects to every spine, leaf-to-leaf traffic crosses exactly one spine, hop count and latency stay constant, and capacity scales by adding spines.
Availability arithmetic
Components in series multiply: A = A1 × A2. Redundant paths in parallel combine as A = 1 − (1−A1)(1−A2). N+1 provides one spare; 2N duplicates everything and also removes shared maintenance outages. A redundant pair sharing one PSU, one uplink or one cooling unit is still a single failure domain.
Physical and power rules
Copper: a 100 m channel is 90 m of permanent link plus 10 m of patch cords total; Cat6A is needed for 10GBASE-T at the full 100 m. OM4 supports 10GBASE-SR to 400 m; OS2 carries building-to-building links. Power over Ethernet is a budget, not a per-port promise: 802.3af supplies 15.4 W at the PSE (12.95 W at the device), 802.3at 30 W (25.5 W), 802.3bt Type 3 60 W and Type 4 90 W (71.3 W at the device).
Worked example
A design has one core switch (A = 0.999), two parallel distribution switches (0.999 each) and one access switch (0.999). The series product is 0.999 × (1 − 0.001²) × 0.999 ≈ 0.998. Adding a second core raises it to roughly 0.999999 for about the same spend as duplicating distribution again, because the single core dominates the remaining risk. Argue for redundancy where the weakest series element sits, not where the diagram looks symmetric.
Traps that cost marks
Equating redundancy with resilience while ignoring shared power, upstream links or the management plane. Assuming a full mesh is always better - it multiplies cost and routing complexity. Confusing collision domains with broadcast domains. Forgetting that a switch stack is one control plane and therefore one failure domain.