Redundant server power supplies exist so the machine survives losing one of them. That requirement sets the sizing: each unit must be capable of carrying the entire server by itself. A pair of units is therefore sized for one server’s worth of load, not two, and treating the pair as combined capacity is how people build systems that fail at the first fault.
What redundancy means arithmetically
The common shorthand describes how many units are needed to run the load and how many spares exist beyond that. A configuration with one unit required and one spare survives a single failure. A configuration with two units required and one spare survives a single failure but not two. The number required is determined by the load and by each unit’s rating, and the spare adds no usable capacity because its entire purpose is to be available when a peer disappears. If a server’s load grows until two units are genuinely needed to carry it, a two-unit machine has quietly stopped being redundant and nobody receives a notification. Management interfaces normally expose the state of each unit, its input voltage, and its output load, and configuring alerts on those is what turns a silent failure into a ticket.
How load sharing affects efficiency
In normal operation the pair usually shares the load, so each unit runs at roughly half the server’s demand. Conversion efficiency in a switching supply is not flat across its range, and half load often sits in a reasonably good part of the curve, but a lightly loaded server with two large supplies can have each unit running well down toward the poor end. Some platforms offer a mode that keeps one unit active and holds the other in a low-power standby, waking it if the active unit fails. That improves efficiency at the cost of a brief transfer, and whether it suits your availability requirement is a decision to make deliberately rather than by default. Efficiency also varies with input voltage, and equipment fed at the higher voltages common in facility distribution generally converts more efficiently than the same unit on a lower-voltage supply.
The mistake that quietly removes the redundancy
Two power supplies fed from the same circuit protect against a supply failure and nothing else. If both cords land on the same rack distribution unit, the same breaker, or the same building feed, then a fault anywhere upstream takes both at once. This is the single most common defeat of redundancy in practice, and it usually happens during a rushed install or a rack reorganization rather than in the original design. The fix is discipline about where each cord goes and a documented record of it, verified by physically tracing cords rather than trusting a diagram drawn two years ago. A pair that has run together since installation shares an operating history and a thermal environment, so a failure in one is a reasonable prompt to plan replacement of the other rather than to assume it is fine.
Testing that the redundancy actually works
Redundancy that has never been exercised is a hypothesis. Test it on a schedule, during a maintenance window, by removing one feed at a time and confirming the machine continues without interruption and that the management system reports the event. A surprising proportion of first tests reveal something: a supply that was never seated fully, a failed unit that had been alarming into a monitoring system nobody watches, or a second feed that turns out to originate from the same source after all. Test both directions, since a fault in the path that carries the load behaves differently from a fault in the idle path. Firmware on the supplies themselves sometimes receives updates, and those updates occasionally change load sharing or reporting behavior, so include them in the platform’s maintenance planning.
Hot swap and the hazards around it
Redundant units are generally designed for replacement while the machine runs, which is much of their value. That does not make the work casual. Rack power distribution equipment is energized at mains voltage, the cords carry substantial current, and pulling the wrong unit on a machine whose peer has already failed drops the server. Identify the failed unit positively from the management interface and from its own indicator before touching anything, and confirm the surviving unit is healthy first. Work inside energized equipment cabinets belongs to qualified personnel following the employer’s electrical safety program, not to whoever is nearest the rack. Blanking plates and correct airflow matter here too, since a supply starved of cool air ages faster and its failure will be attributed to bad luck rather than to a missing panel.
What to check before you act
- Confirm one unit alone can carry the server’s full measured load.
- Physically trace each cord to prove the two feeds are independent.
- Test failover on a schedule rather than assuming it works.
- Positively identify the failed unit before removing anything.
Common questions
Can I run a server on one supply if the other has failed?
Yes, provided the remaining unit is rated for the full load, which is the whole point of the design. Replace the failed unit promptly, since you now have no spare.
Does using both supplies halve the wear on each?
Load sharing does reduce the thermal stress on each unit, but both are powered and aging. Treat them as a pair of the same age.
The practical takeaway
Size each supply for the whole server, route the two cords to genuinely independent sources, and test the failover on a schedule. Redundancy that has not been traced and tested is documentation rather than protection.
References and further reading
- NFPA 70E: electrical safety in the workplace
- OSHA 1910.304: wiring design and protection requirements
- Department of Energy: computing equipment efficiency guidance
Numerical scenarios are illustrative unless identified otherwise. Follow the exact product instructions; component ratings and local installation requirements can differ.
