Power is the one thing every device in the rack shares. A tripped breaker, a failed UPS or a PDU with a bad outlet takes everything down at once, usually on a weekend. The layout of the power path decides whether that failure is a log entry or an outage.
The goal is simple: no single point of failure between the wall and the server. That means two PDUs, two feeds where the building allows it, dual power supplies plugged into different PDUs, and a UPS with enough capacity and runtime to ride through the short outages and shut things down cleanly in the long ones.
Two PDUs, two sides, two feeds
Mount two vertical PDUs in the rear of the rack, one on each side. Call them A and B. Every device with dual power supplies plugs one supply into A and the other into B. A device with a single supply plugs into one PDU, and you accept the risk or add a small automatic transfer switch in front of it.
Where the building has two circuits available, put the A PDU on one and the B PDU on the other, from different breakers and ideally different panels. Where there is only one circuit, both PDUs still hang from it, but the split still protects you from a PDU failure and from a fault in one power cord. Label each PDU and each outlet with the feed name so nobody plugs both supplies of a server into the same side.
- PDU A on the left rear, PDU B on the right rear, labelled at top and bottom.
- Dual-PSU devices: one cord to A, one cord to B, cords labelled with the device name.
- Single-PSU devices: pick one side, or add an automatic transfer switch.
- Different breakers for A and B where the building allows; different panels if you can get them.
Do not overload either side
The trap with A/B feeds is that each side must be able to carry the whole rack alone. If side A fails, every dual-supply device draws its full power from side B. So each PDU and its breaker must be sized for the total load of the rack, not half of it. In practice that means keeping the normal running load on each side well below half of the breaker's continuous rating.
Read the load from the PDU's display or web interface and write it down at commissioning and after every change. Breakers are derated to 80 percent for continuous load, so a 20 amp circuit should carry no more than 16 amps, and with A/B redundancy each side should sit at or below half of that so a failover does not trip the surviving breaker.
- Find the continuous rating of each circuit: breaker amps times 0.8.
- Measure the actual running load on each PDU under normal conditions.
- Confirm that the combined load of both sides fits within one side's continuous rating.
- Test it: with everything running, unplug PDU A's feed and watch PDU B's reading. Then swap.
Sizing the UPS
Two numbers matter: capacity and runtime. Capacity is how much load the UPS can carry, given in VA and in watts; use the watt figure and compare it to your measured load, not the sum of the nameplate ratings on the power supplies, which are far higher than what the servers actually draw. Leave headroom for growth and for the fact that a UPS running near its limit has a shorter battery life.
Runtime is how long the batteries hold that load. For a small office the honest target is not hours; it is long enough to ride through blips of a few minutes and, in a longer outage, to shut the servers down cleanly. That means a network management card in the UPS and a shutdown agent on the servers, tested. A UPS without that link just delays the crash. If the site needs real runtime, a generator is the answer.
- Size from measured watts with headroom, not nameplate ratings.
- Runtime target: ride through short blips plus enough time for a clean shutdown of everything.
- Network management card in the UPS, shutdown agent on every server and hypervisor, tested at least once a year.
Cords, connectors and the things people forget
The PDU outlets and the server cords must match. Most rack PDUs use C13 and C19 outlets; most servers ship with cords that fit a wall socket, which are useless in a rack. Order C13-to-C14 cords in the lengths you need, and C19-to-C20 for high-draw devices such as large UPS units. Outlet retention clips stop a cord from walking out when a neighbouring server is slid out.
Confirm the wall side too. A rack that needs a 30 amp feed will not run from a standard office outlet, and the electrician needs the receptacle type from the UPS or PDU spec sheet before the gear arrives. Finally, put the switch that the UPS management card uses on the UPS itself, or the shutdown signal will never reach the servers when the power fails.
Frequently asked questions
Is A/B worth it if there is only one circuit?
Yes. It still protects against a failed PDU, a bad outlet or a damaged cord, and it makes adding a second circuit later a matter of moving one plug.
Can I plug the second PSU into the wall instead of a UPS?
It is better than nothing, but a wall outlet on the same building circuit as the UPS input offers no protection from an outage. Two UPS units, or one UPS feeding both PDUs, is the usual answer.
Takeaway
Two PDUs on opposite sides, dual power supplies split across them, feeds from separate breakers where possible, and each side sized to carry the whole rack alone. Size the UPS from measured watts, plan its runtime around a clean shutdown, and test the whole chain. Power is the layer that fails silently until it fails completely.