Ethernet and Cabling: The Physical Layer

Every road network has the same three questions: what kind of road, how far can it go, and what happens when the pavement is damaged. A narrow city street can't carry highway traffic; a highway stretched too far needs a stronger car; and a single pothole in the right place slows every truck behind it. The cables under your floor and inside your walls ask the exact same questions — some carry short bursts to a desk a few meters away, others carry a building's worth of traffic for kilometers, and one dirty connection can halve the speed of everything passing through. The maddening part: a bad cable rarely announces itself. It just quietly makes everything slower, and the damage shows up as a number on a screen.

Here's the puzzle.

Scenario. You're the engineer signing off on a campus upgrade: a new fiber run between Building A and Building B, 10.0 km apart. The re-termination crew is on site today only — if the link's light math doesn't add up, you must tell them which component to fix before they leave. Sign off wrong, and the link flaps under load for months. Here are the exhibits.

Given artifacts. The run (hover any element for details):

Exhibit A — Transceivers (SFP+ 10GBASE-LR, one at each end)
  Transmit power (minimum):  -3 dBm
  Receiver sensitivity:      -11 dBm

Exhibit B — Cable plant (single-mode, 10.0 km building-to-building)
  Cable loss:    0.4 dB per km
  Connectors:    6 at 0.75 dB each
  Splices:       2 at 0.25 dB each

Your task: Deliver three things: (1) the power budget in dB, (2) a PASS/FAIL verdict on the link, and (3) if it fails, the one component you'd change — and the recomputed numbers proving the fix works.

Workspace: Analyze-and-answer — type your budget, verdict, and component below, then check the commitment box to reveal the worked answer.

☐ I've attempted this challenge and thought it through. (Checking reveals the worked answer in S7 — honor system: the page hides it until you commit.)

Copper: the city streets

Copper cable carries data as electricity, and it's the default for short runs — desks, offices, wiring closets. The categories are really distance-and-speed ratings:

• Cat5e (the Category 5 enhanced cable — the older office standard): 1 Gbps up to 100 meters. Fine for a desk; hopeless for a data hall.
• Cat6: 1 Gbps to 100 m, and 10 Gbps but only to about 55 m — the faster signal fades sooner.
• Cat6a (the augmented version): 10 Gbps all the way to 100 m, thanks to thicker shielding against interference.

Inside the jacket are twisted pairs of copper wires — the twisting cancels out electrical noise from neighboring cables. Push any copper past its rating and you don't get a clean failure; you get errors, retransmissions, and a link that looks up but performs terribly.

Why this matters for the challenge: the challenge run is 10 km — roughly a hundred times copper's limit. That's why it's fiber, and why the math is about light, not electricity.

Fiber: the highways

Fiber carries data as pulses of light through glass, and it owns everything long:

• Single-mode fiber (light travels down one path — one mode — straight down the core): kilometers of reach, the choice for campus and city links. Thinner core, laser light source, pricier electronics.
• Multi-mode fiber (light bounces down many paths at once): cheaper, but the bouncing smears the signal — hundreds of meters at most. The choice inside a single building or data hall.

Light doesn't care about electrical interference, lightning, or motors — a huge practical win on factory floors and between buildings. The tradeoff: glass is fragile, connectors must be spotless (a speck of dust is a boulder to a light beam), and every connection steals a little light.

Why this matters for the challenge: the exhibits describe single-mode — the right call for 10 km — and every connector and splice on that run takes its dB toll.

The plug-in engines

The transceiver (the plug-in module that converts electricity to light and back — the engine at each end of a fiber run) comes in standard sizes:

• SFP (small form-factor pluggable): 1 Gbps — the older workhorse.
• SFP+ (the plus = 10 Gbps version): same size, ten times the speed — what the challenge run uses.
• QSFP (quad small form-factor pluggable): 40 or 100 Gbps — four lanes in one module, the data-center standard.

Each transceiver publishes two numbers that define your budget: how much light it sends (transmit power) and the faintest signal the far end can still read (receiver sensitivity). The gap between them is everything you have to spend on the cable plant.

Why this matters for the challenge: Exhibit A is these two numbers — the entire "money" side of your budget.

Speed, duplex, and the auto-handshake

Two more agreements every link needs, usually handled automatically:

• Speed — 10, 100, 1000 Mbps and up. Both ends must agree, or they can't understand each other at all.
• Duplex — half duplex (take turns: talk or listen, like a walkie-talkie) versus full duplex (talk and listen at once, like a phone call). Modern links are full duplex.

Autonegotiation (the quick handshake where the two ends agree on speed and duplex by themselves) almost always gets this right — which is why you should usually leave it on. The classic failure: someone pins one end to full duplex while the other end still auto-negotiates. The auto end, hearing no negotiation, falls back to half duplex — and now you have a duplex mismatch: the link stays up, but one side talks over the other, producing collisions, errors, and miserable speed. It looks like a "slow network," not a broken one — which is exactly why it's so confusing.

Why this matters for the challenge: not directly — the challenge is pure physics — but the first thing you'll check on any copper link that "works but slowly" is this agreement.

Decibels: the budget math

dB (the decibel — a compact way to add up gains and losses instead of multiplying tiny fractions) is the language of light budgets. Optical power is usually quoted in dBm (decibels relative to one milliwatt — a scale where 0 dBm = 1 mW, and the numbers go negative for the faint signals in fiber).

You only need three moves:

1. Power budget = transmit power − receiver sensitivity. With −3 dBm out and −11 dBm minimum readable: (−3) − (−11) = 8 dB to spend. (Subtracting a negative adds — the most common arithmetic slip in this business.)
2. Plant loss = (km × dB/km) + (connectors × dB each) + (splices × dB each). Just add the tolls.
3. Margin = budget − plant loss. Positive means PASS with room to spare; negative means FAIL — the receiver is being asked to read a signal fainter than its limit.

A thin margin (under ~3 dB) passes on paper but leaves no room for aging, dirt, or a bad splice day — engineers like cushion.

Why this matters for the challenge: this is the entire deliverable — budget, verdict, and the fix, all in dB.

Platform: Cisco IOS / IOS-XE

! PLAIN-ENGLISH: Ask the switch what this port is really doing — negotiated speed, duplex, and error counters.
show interfaces GigabitEthernet1/0/1

The two lines that matter in the output (the rest is context):

GigabitEthernet1/0/1 is up, line protocol is up
  Full-duplex, 1000Mb/s, media type is 10/100/1000BaseTX
  0 input errors, 15432 CRC, 0 frame, 0 overrun, 0 ignored

Full-duplex, 1000Mb/s is the autonegotiation result — what the two ends agreed on. The 15432 CRC errors with zero input errors is the classic fingerprint of a duplex mismatch or a sick cable: frames arrive damaged. (On a fiber port you'd also check the light levels — but no command fixes a budget that doesn't add up.)

! PLAIN-ENGLISH: Stop auto-guessing and pin the port to 1 Gbps, full duplex (safe only if the other end is pinned identically).
interface GigabitEthernet1/0/1
 speed 1000
 duplex full

⚠️ Remove this, break that: pinning duplex full on only one end creates the classic mismatch — the auto side falls back to half duplex, the link stays up, and everything crawls with collisions and CRC errors. ⚠️ Remove this, break that: speed 1000 disables autonegotiation on this port — if the other end is still auto-negotiating, it may guess the speed wrong and the link won't come up at all. ⚠️ Remove this, break that: some faults aren't config lines at all — they're physics. No command repairs a link whose light budget doesn't add up; that one is fixed with a cleaning kit, not a keyboard.

• Match the cable to the distance: copper for meters, single-mode fiber for kilometers — the wrong medium fails no matter how good the config.
• A duplex mismatch (one side half, one full) doesn't kill the link — it cripples it with collisions and errors, which is why it's so confusing.
• Leave autonegotiation on unless both ends are pinned identically; one-sided pinning is the classic mismatch.
• A fiber link passes only if total plant loss (cable + connectors + splices) stays under the transceiver's power budget — do the dB math before you sign off.
• When a budget fails, attack the biggest loss line item first — usually the connectors, the cheapest part to fix.

Next: How a Switch Learns: The MAC Table — You've mastered the physical road; next you'd learn how a switch memorizes every address on it — but this is where the free taste ends and the paid track begins. Module 2 continues the journey for subscribers.