Network Design Fundamentals and Business Alignment
This domain tests whether you can convert a business outcome into a defensible technical design: eliciting requirements, quantifying them, and proving the design satisfies them at an acceptable cost.
From business driver to testable requirement
A business driver states an outcome ("order capture must survive the loss of a data center"). A requirement is something you can test ("failover completes in ≤ 30 s with zero committed transactions lost"). Strong answers convert every driver into a measurable requirement and map it to a specific design decision - topology, redundancy model, protocol timers, budget line.
- Functional vs non-functional: capacity, latency, availability, security, regulatory, cost.
- Constraints: budget, existing staff skills, legacy platforms, schedule, contractual SLAs.
- Scope: which applications, flows, sites and user populations are actually in scope.
- Operating model: in-house vs managed, who owns which domain, change windows.
Availability and downtime budgets
Serial availability multiplies: A_total = A1 × A2 × … Parallel (redundant) paths give A = 1 − (1 − A1)(1 − A2). Convert percentages to time before arguing: 99.9% ≈ 8.76 h/yr, 99.95% ≈ 4.38 h/yr, 99.99% ≈ 52.6 min/yr, 99.999% ≈ 5.26 min/yr. Availability also equals MTBF / (MTBF + MTTR), so cutting MTTR with fast convergence or automation is usually cheaper than buying more reliable hardware.
Worked example
A requirement says the order path may lose no more than 60 minutes per year (≈ 99.988%). The path is core switch → firewall → application. Each device is rated 99.99% (52.6 min/yr). Three in series: 0.9999³ ≈ 0.9997, about 157 min/yr - the design fails. Pair the firewall (two 99.99% units in parallel give 1 − 0.0001² ≈ 99.999999%) and pair the core, and the arithmetic passes easily. But the answer is not finished: if both firewalls sit on one power feed in one rack, the parallel calculation is fiction. Redundancy must remove common cause - power, rack, building entry fiber, the same upstream WAN link, the same DNS or AAA server. The real weakest link is often those shared services, not the routers.
Traps that cost marks
- Accepting "five nines" without asking which service, over what window, and whether it is measured per transaction or per minute.
- Computing availability for network devices only, while ignoring DNS, DHCP, AAA and WAN carried as serial dependencies.
- Adding redundancy the business cannot afford to operate; complexity has a recurring cost and its own outage rate.
- Skipping the growth model, so the design is overtaken by capacity within the depreciation cycle.