Trace Both Power Feeds Before Calling a Rack Redundant

A server with two power supplies can still lose both when one breaker trips. The connectors are visible from the aisle. The shared dependency may be several rooms away. For a redundancy review, start with one server and trace both paths as far upstream as the facility can document.

Draw the actual paths

Label the server inlets, rack PDUs, circuits and upstream distribution. If the two cords terminate on different outlets of the same PDU, the rack has separate plugs but a common failure point. Separate PDUs may still share a breaker, UPS or maintenance bypass. Mark what is independently verified and what the provider has yet to confirm.

Power is only one dependency. A storage shelf, top-of-rack switch or cooling unit can take down a service even when the server itself stays powered. Track those dependencies in the service diagram instead of inferring application availability from a facility label.

Do the failover arithmetic

Suppose a rack normally draws 3 kW from A and 3 kW from B. If all dual-fed devices transfer their load, the surviving feed may need to carry roughly 6 kW. Verify actual transfer behavior, power supply capacity, phase loading and the facility's allowable circuit load. Never assume the nominal breaker rating is the permissible continuous operating load.

Include single-cord equipment in the review. A transfer device may help, but it introduces its own operating limits and failure modes.

Verify safely

Use approved maintenance procedures with the facility operator and qualified staff. Record expected alarms, load on the surviving path and a stop condition before a planned test. A production plug-pull is not a substitute for engineering review. The useful result is a dated diagram and measured failover margin that the next operator can understand.

Further reading

Uptime Institute: Data center tiers

Related guide: VM Snapshots and Backups Solve Different Problems.