Racks, Power, and Cooling — The Physical World
Hyperscaler Network Engineer · Module 1: Data Centers from Zero
Lesson 3 of 8
Prerequisites: What Is a Data Center, Really?
What you'll be able to do: Read a rack elevation diagram, explain why every rack needs two power feeds and a cold aisle, and spot the physical faults hiding behind a "network outage."
A data center is a city. The racks are the skyscrapers, electricity is the water grid, and cooling is the air the whole city breathes — because tens of thousands of packed machines generate heat like an oven left on. The network — the part that moves information around — gets all the glory, but it is helpless without the physical city underneath it: if the power blinks, the computers stop; if the cooling fails, they overheat and shut themselves down within minutes to avoid melting. Most "network outages" that make the news turn out, on inspection, to be power or cooling failures wearing a network costume. So before we talk about how computers talk to each other, we have to walk the streets: the racks they live in, the two power feeds that keep them alive, and the rivers of cold air that keep them breathing.
Here's the puzzle.
Scenario. You are the night-shift facilities tech. A ticket says Rack R-17 "looks wrong" after yesterday's maintenance visit. Nobody else is on site until morning, and this rack holds part of the company's login system — if it goes dark, nobody can sign in. Your only tool is the rack's elevation diagram below. Find the physical risks before they find you.
Given artifacts. The elevation diagram of Rack R-17 (hover each element for details). Read it like a building floor plan: U42 is the top slot, U1 the bottom. What's given: every device, its rack unit, its power plugs, and which way it faces. What's not given: anyone to ask — the diagram is all you have.
Your task: Name the TWO physical risks in Rack R-17, state the rack unit each one lives at, and give the one-line fix for each.
Workspace (analyze-and-answer): Two short entries — "Risk 1: what + which U + fix" and "Risk 2: what + which U + fix." The page records your answers when you hit Commit; nothing is graded.
Hint ladder:
Hint 1 — where to look
A rack has exactly two life-support systems: power and air. Check each one separately — first follow the electricity from the wall to the servers, then follow the air through the switch.Hint 2 — what to compare
Count the power paths: every device should have TWO live paths (feed A and feed B). Find the device whose second path is a dead end. Then check which way the switch breathes — every device must inhale from the cold aisle.Hint 3 — the mechanism
A power distribution unit with no feed cable is just a fancy power strip with no power behind it: everything "dual-plugged" into it is secretly running on one feed. And a switch is a wind tunnel with a direction — intake must face cold air and exhaust must face hot; flip it around and it air-conditions itself with its own hot exhaust until it gives up.Commitment ritual: When you have thought it through, check the box:
- [ ] I've attempted this challenge and thought it through.
Checking it reveals the worked answer in S7. (Honor system — the page hides the answer until you commit.)
Checking the box reveals the worked answer in S7 below. Returning learners stay unlocked.
The rack: a 42-unit skyscraper
Servers don't sit on desks — they live in racks (in plain English: standard steel frames, 19 inches wide, that hold equipment like trays in an oven). Space inside a rack is measured in rack units, abbreviated U (in plain English: one U is about 1.75 inches of vertical height; a full-size rack is 42U tall). A rack elevation diagram (in plain English: a side-view floor plan of one rack, with U1 at the bottom and U42 at the top, showing what lives in each slot) is how engineers talk about physical layout without walking to the building. Everything is measured in U because space is money: a taller server (2U, 4U) rents more of the rack.
Why this matters for the challenge: the challenge is literally a rack elevation diagram — U numbers are the addresses where the risks live, so read them like street numbers.
Power: two feeds and the PDU
Electricity reaches a rack through a power feed (in plain English: an independent path of electricity from the building to the rack). Every serious rack has two: feed A and feed B, coming from separate breakers and ideally separate sources. Power then spreads through a PDU — a power distribution unit (in plain English: the rack's industrial power strip) — one PDU per feed. Servers built for data centers have two power cords, one plugged into each PDU. The mechanism is simple: if feed A dies — a tripped breaker, a failed utility line, a backhoe through a cable — everything keeps running on feed B without a blink. But redundancy only works if both paths are actually live. An unplugged cable, a switched-off breaker, a PDU nobody verified — any of these quietly converts your two feeds back into one, and you won't find out until the remaining feed fails.
Why this matters for the challenge: one of the rack's risks is a power path that looks complete but isn't — trace every plug to a live feed, and trust indicator lights over assumptions.
Cooling: hot aisles, cold aisles, and the CRAC
Tens of thousands of machines in one building generate heat like an oven left on, so data centers choreograph the air. Racks are lined up front-to-front and back-to-back: the aisle the fronts face is the cold aisle (in plain English: the corridor into which chilled air is pushed), and the aisle the backs face is the hot aisle (in plain English: the corridor where hot exhaust air collects). Giant CRAC units — computer-room air conditioners (in plain English: industrial chillers that cool the building's air and push it into the cold aisles) — keep the cycle going. The mechanism inside each machine: servers are wind tunnels — fans pull cold air in through the front, it absorbs heat from the electronics, and hot air blows out the back. Mount a server or switch facing the wrong way and it inhales its neighbors' exhaust, gets hotter, spins its fans harder, and eventually shuts itself down to avoid melting.
Why this matters for the challenge: the second risk is an airflow direction problem — check which aisle each device breathes from, not just whether it's plugged in.
The outage that was never the network's fault
Picture this: on a hot summer afternoon, a building's cooling plant fails. Within minutes, servers start inhaling warmer and warmer air. One by one they hit their thermal limits and shut themselves down to protect their electronics. To everyone watching the dashboards, it looks exactly like a network outage — nothing responds, everything is "down," the network team gets paged. But no cable failed and no switch broke; the machines simply refused to cook themselves. The same costume party happens with power: a rack silently running on one feed looks perfectly healthy right up until that feed blinks, and then the whole rack goes dark at once. The lesson veterans learn the hard way: when an entire rack or row dies together, suspect the building before you suspect the network.
Why this matters for the challenge: Rack R-17 is wearing both costumes at once — one risk in the power system, one in the airflow. Check both life-support systems and you'll find both.
- Step 1 of 5: The CRAC unit pushes chilled air down into the cold aisle between the two rack rows.
- Step 2 of 5: Every server inhales that cold air through its front — blue arrows flowing into the racks.
- Step 3 of 5: Heated air blows out the backs into the hot aisles, where red arrows carry it up and away to be chilled again.
- Step 4 of 5: Now the wrong-way server: mounted backwards, it inhales its neighbors' hot exhaust instead of cold air.
- Step 5 of 5: Watch its temperature bar climb — a backwards device air-conditions itself with hot air until it shuts down to survive. Direction is everything.
🔒 Revealed after the commitment ritual in S2 — attempt the challenge first. (Honor system: the page hides this until you check the box.)
Risk 1 — the power path (lives at U40). Follow the electricity: PDU-A at U41 is plugged into feed A with a green light — live. PDU-B at U40 has its feed-B cable missing, indicator dark. That makes PDU-B a power strip with nothing behind it, which means every server's second cord (S1–S4 at U38–U39, S5–S8 at U10–U12) is a dead end. The whole rack is secretly running on feed A alone. One tripped breaker on feed A and the login system goes dark instantly — at 3 a.m., with nobody on site. Fix: plug the feed-B cable back into PDU-B and confirm its indicator light goes green before morning.
Risk 2 — the airflow (lives at U20). Check which way each device breathes: SW-1 is mounted backwards, its air intake facing the hot aisle. It's inhaling hot exhaust air, so it will run hotter and hotter until it shuts itself down to avoid melting — and when the switch dies, every server plugged into it loses its network, which looks exactly like a network outage. Fix: remount SW-1 with its intake facing the cold aisle.
Wrong turns, named. You might have blamed the servers at U38–U39 — but they're dual-plugged correctly; the fault is upstream at the PDU, and blaming the victims is the classic misread. You might have suspected the patch panel at U42 — it's passive wiring with no power and no fans; it can't fail this way. You might have dismissed one backwards switch as harmless because "it's just one box" — but its shutdown takes its connected servers' network with it, and heat doesn't stay politely inside one slot.
Verify it worked: Walk the rack once more with two questions. One — does every device have two live power paths? Check both PDU indicator lights, not just the plugs. Two — does every device breathe from the cold aisle? Stand in front of each one: you should feel cool air on your face. Two green lights and cool air on every face means the rack is healthy.
Check yourself — nothing here is graded. Wrong answers are the useful ones; each explains why.
Question 1. What is a rack unit (U)?
Question 2. Why does a serious rack have two power feeds, A and B?
Question 3. In a hot-aisle / cold-aisle layout, which way should a server's air intake face?
Question 4. Every server in a rack goes dark at 2 a.m. The network team reports the switches are fine — 'they just have nothing to talk to.' Facilities finds one tripped breaker. What design would have prevented the outage?
Question 5. Put the cooling airflow in order:
- A rack is real estate measured in U — 42U tall, U1 at the bottom — and reading an elevation diagram is reading the building's floor plan.
- Every device deserves two live power paths (feed A plus feed B); redundancy you never verify is redundancy you don't have — an unplugged cable is the same as no second feed.
- Air has a direction: cold aisles feed the fronts, hot aisles collect the backs — face a device the wrong way and it will overheat on its own exhaust.
- Servers protect themselves from heat by shutting down, which looks exactly like a network outage to everyone watching the dashboards.
- When a whole rack or row dies together, check power and cooling first: the cheapest failures wear the most expensive costumes.
Next: Servers, Switches, and Cables — The Hardware Zoo — The building is alive with power and cold air — now meet the residents: the boxes that do the thinking, the boxes that direct the traffic, and how to pick the right cable so a short link doesn't become an expensive lesson.