01 / TWO COOLING PATHS
Count what survives, not only what is installed
A resilient design may have multiple cooling units, CDUs, pumps, liquid loops, facility-water paths, or heat-rejection resources. The useful question is not just “How many components are there?” It is: If Path A disappears, can Path B still remove the required heat?
REDUNDANCY ≠ DUPLICATION
Two components are not yet a resilient service
Duplication
“There are two units.”
Counts equipment in the design.Redundancy
“Required thermal service continues after the intended failure.”
Tests the surviving system against the real load.Two CDUs connected to one shared pump or header may retain a common dependency. Two units powered from one vulnerable source may fail together. Two liquid paths that converge on one critical facility-water component may not be independent at that boundary. A shared component is not automatically a defect; identify it and understand its failure impact.
N · N+1 · 2N
Capacity labels describe a defined level
At beginner level, these labels describe capacity relative to a required load at some chosen boundary. They do not, by themselves, prove that a whole facility will survive a failure.
- N
- Capacity required to support the intended load.
- N+1
- Required capacity plus one additional unit or component at the defined level.
- 2N
- Two full-capacity systems or paths at the defined level.
Topology, independence, surviving capacity, controls, and maintenance state still matter. Always ask what boundary N+1 or 2N applies to: a pump group, a cooling unit, a CDU, or an end-to-end path to heat rejection.
02 / ONE PATH UNAVAILABLE
Installed capacity is not surviving capacity
Consider a simplified example. IT heat load is 220 kW. Path A can provide 150 kW, and Path B can provide 150 kW. Both together provide 300 kW installed capacity. If Path A fails, only 150 kW remains—less than the 220 kW load.
150 kW < 220 kWRemaining capacity is below this example load.
Normal
- Path A — available · 150 kW
- Path B — available · 150 kW
300 kW availableEnough for the 220 kW example load
Path A failed or isolated
- Path A — unavailable
- Path B — available · 150 kW
150 kW remainsBelow the 220 kW example load
Installed capacity ≠ surviving capacity. Redundancy is useful only when what survives can support the load.
PLANNED OUTAGE · SAME QUESTION
Maintenance is also a resilience test
Redundancy is not only for unexpected faults. Ask whether one CDU, pump, cooling unit, or path can be isolated for planned maintenance while the remaining infrastructure supports the current IT heat load. Maintainability is part of resilience thinking.
03 / LOOK FOR THE HIDDEN SHARED DEPENDENCY
Two paths can converge on one point
Paths that look separate at the rack may share a supply or return header, pump, control system, power source, CDU, heat exchanger, facility-water loop, chiller or dry cooler, or physical route. One shared component can affect both apparent paths, depending on how the system is designed and operated.
Physical independence
Do paths avoid inappropriate shared physical infrastructure?
Functional independence
Can the surviving path operate, control, distribute, and remove enough heat after a failure?
A path may look physically separate yet share power, controls, sensors, pumps, valves, plant capacity, or another operating dependency. Trace the path and determine what the shared element does before deciding whether it is acceptable.
CONNECT TO LESSONS 04 AND 05
Surviving capacity must reach the affected racks
Even when capacity remaining is numerically sufficient—total capacity ≥ IT heat load—the cooling must travel through a usable path and reach the right place. Airflow or hydraulic distribution, local rack density, valves, pump limits, blocked paths, poor containment, and hotspots can reduce what the affected equipment actually receives.
This is a beginner mental model, not an engineering equation. See Airflow & Containment · Lesson 04 for delivery paths and Cooling Capacity & Failure · Lesson 05 for the difference between installed, available, and local cooling.
TRACE THE DEPENDENCIES
Five failure states, five questions
One cooling unit fails
How much usable capacity remains?
One CDU fails
Is another path available to the affected racks?
One pump fails
Is liquid flow maintained by another pump or path?
Shared facility-water header fails
Do both apparent paths depend on that header?
One path is under maintenance
Can the remaining path support the current IT heat load?
Outcomes depend on topology, controls, remaining capacity, and site conditions. Ask the question for the actual failure boundary instead of assuming every site responds the same way.
THE COOLING FAILURE CHECKLIST
When a cooling component fails, ask:
- What failed?
- What cooling loads depend on it?
- What alternate cooling path remains?
- Is that path actually independent?
- How much cooling capacity remains?
- Can the remaining capacity reach the affected racks?
- Are flow, airflow, temperature, and controls still acceptable?
- Are there hidden shared dependencies?
- Does the required service remain within acceptable thermal conditions?
ONE ENGINEERING QUESTION · THREE SYSTEMS
Power, network, and cooling follow the same logic
Power redundancy
Can another power path still support the load?
A/B Power & RedundancyNetwork redundancy
Can another network path still reach the service?
Network RedundancyCooling redundancy
Can another cooling path still remove the heat?
CDU & Liquid Cooling Loop · Lesson 07PUT FAILURE THINKING INTO PRACTICE
Now remove a cooling dependency
Use the existing labs for their defined scope. CDU Failure Diagnosis explores CDU-A loss, diagnosis, restoration, and cold-plate/service recovery. AI Rack & Liquid Cooling lets you inspect a built liquid-cooling branch. N-1 Resilience evaluates its defined UPS/PDU electrical failure cases; it does not certify cooling-component N-1 or redundant CDUs serving the same cold plate.
These labs use the current DataCenterTwin model and are educational evidence, not facility certification. The N-1 lesson is explicitly scoped to UPS/PDU loss.
COOLING 101 · COMPLETE
You now understand the basic data-center cooling path
- IT power
- Heat
- Air or liquid
- Cooling equipment
- Distribution / containment
- CDU / facility cooling
- Heat rejection
The deeper engineering question is always:
“What happens when one part fails?”
Across Power 101, Network 101, and Cooling 101, follow the same method: trace the path, identify shared dependencies, remove one component, find what remains, check capacity, and verify service impact.
THREE THINGS TO REMEMBER
Keep the failure boundary in view
- Two cooling components do not automatically mean resilient cooling.
- The surviving path must be independent and have enough usable capacity.
- Hidden shared dependencies can defeat apparent redundancy.
PASS IT ON