Two labelled power feeds can be useful, but the labels alone do not prove independence. To reason about an A/B arrangement, trace each path from the equipment back to its sources and mark every shared dependency. Then check the actual operating state and capacity of the remaining path.
Follow the chain before judging resilience
In the ATS-01 teaching model, path A runs from the utility/source-selection boundary through shared upstream distribution, UPS-A, PDU-A and a rack distribution unit to a device supply. Path B has UPS-B and PDU-B, but it shares the upstream source-selection and distribution boundary.
The simplified drawing intentionally leaves out protection coordination, earthing, switching details and interlocks. It cannot establish a real facility classification or safe switching plan. Its purpose is to expose a reasoning trap: separate downstream cabinets can still depend on the same upstream asset.
Availability and redundancy are different questions
A device can remain online after losing one supply input. The service may be available while the remaining resilience is reduced. A useful handover reports both facts. “Service online, B input unavailable, A path indication healthy, independence not yet confirmed” is more informative than “all normal.”
Likewise, a UPS online indication does not establish how long stored energy will support the load. Duration depends on the particular system, condition, load and valid test information. A runtime estimate should be interpreted with its assumptions, age and uncertainty.
Capacity needs a defined boundary
For a fictional planning exercise, suppose eight racks total 32 kW and each path has a teaching budget of 40 kW. If one path must carry the entire IT load, the arithmetic remaining headroom is 40 − 32 = 8 kW. That number is not permission to transfer a real load. Real decisions also depend on distribution limits, protection, manufacturer ratings, operating conditions and approved procedures.
Adding both 40 kW path budgets and calling the result 80 kW of usable redundant capacity ignores the failure case. Define whether the objective is normal operation, loss of one path or another scenario before calculating headroom.
Practise tracing a shared failure
Draw two paths using the ATS-01 chain above. Circle the shared components. Now remove UPS-B from your paper model and write what is still unknown. Then remove the shared upstream distribution and describe how the affected scope changes.
Evidence worth comparing
- Current equipment supply telemetry and independent service checks
- Upstream and downstream indications with aligned timestamps
- Approved connection records and asset identifiers
- Recent change history and outstanding maintenance restrictions
- Capacity assumptions at each distribution boundary
Next, try the short sample investigation to practise choosing an information-gathering action before making a change.
Source context
See Eaton’s data centre infrastructure overview for background on power distribution and UPS roles. The topology and arithmetic here are original fictional teaching examples, not an OEM design or an operating procedure.