Power for servers: redundant supplies, UPS and PDUs

Why this matters

A server can have mirrored disks, two network connections and a cluster partner, and still go down because both of its power cords were plugged into the same strip. Power is the dependency underneath every other kind of redundancy, and it is where the most avoidable outages come from.

This lesson follows the power path from the wall to the server: the uninterruptible power supply, the power distribution unit, the connectors, and the server's own power supplies. It ends with the arithmetic that decides whether it all holds up when one part of that path fails.

The lesson

Redundant power supplies, and why the two belong on separate circuits

Most rack-mount servers can take two power supply units (PSUs). In a redundant configuration, often written 1+1, either supply can run the whole server on its own. If one fails, the other carries the load and the server keeps running. The supplies are usually hot-swappable, so the failed one can be replaced without shutting the server down.

Two supplies only give redundancy if they are fed separately. The standard design is A and B feeds: PSU 1 plugs into a PDU on circuit A, PSU 2 into a different PDU on circuit B, ideally from separate UPS units and, in larger sites, from separate electrical panels. Then a tripped breaker, a failed PDU or a UPS fault takes down one feed while the server runs on the other.

Plug both supplies into the same PDU, or two PDUs on the same circuit, and the second supply protects only against the PSU itself failing. The first breaker that trips takes out both. This is one of the most common findings in any review of a server room, and the exam likes it because it looks redundant on paper.

There is a sizing consequence too. When one feed fails, all of the load moves to the other. Each feed must therefore be able to carry the full load of every server connected to it, which in practice means each feed normally runs at no more than half of what it can deliver.

UPS types -- standby, line-interactive and online -- and what each protects

An uninterruptible power supply (UPS) sits between the mains supply and the equipment and keeps it running from batteries when the mains fails. There are three designs, and the difference is what happens to the power the rest of the time.

• A standby (offline) UPS passes mains power straight through and switches to battery only when the mains fails or goes out of tolerance. The switch takes a few milliseconds. It is the cheapest design and suits desktops more than servers.
• A line-interactive UPS adds an automatic voltage regulator that corrects sags and surges without using the battery. It switches to battery for longer or deeper problems. It is common for small server rooms and network cabinets, and it handles the frequent small disturbances well.
• An online, or double-conversion, UPS converts incoming power to DC and back to clean AC all the time, so the equipment always runs from the inverter. When the mains fails there is no transfer at all, and the output is isolated from noise and fluctuations on the incoming supply. It gives the best protection and costs more to buy and run, since the constant conversion wastes some power as heat.

A UPS is not meant to run servers for hours. Its job is to ride through short interruptions and give time either for a generator to start or for servers to shut down cleanly. That second job needs the UPS connected to the servers by USB, serial or a network management card, with shutdown software configured to act when the battery runs low. A UPS without that connection simply runs until it dies and the servers lose power abruptly anyway.

UPS batteries wear out, typically within three to five years. Most units self-test and report a failing battery, and those alerts should go somewhere a person will see them.

PDUs, metered against switched, and balancing load across them

A power distribution unit (PDU) distributes power to the equipment in a rack. In a rack it is usually a vertical strip mounted at the back, with many outlets.

• A basic PDU is a heavy-duty power strip and nothing more.
• A metered PDU shows how much current it is drawing, locally or over the network. This is how you know how close a circuit is to its limit.
• A switched (managed) PDU adds remote control of individual outlets, so a hung device can be power-cycled from anywhere, and outlets can be switched on in sequence after an outage so every server does not draw its startup surge at once.

Each PDU is limited by the circuit feeding it. Balance the load across the available PDUs and circuits rather than filling one first, and keep an eye on the metered readings as equipment is added. In many regions, including North America, electrical practice limits continuous load on a circuit to 80% of its breaker rating, and a planned change that would push a PDU past that needs a new circuit, not an extra plug.

Server power connectors, C13/C14 and C19/C20, and what they carry

Server and PDU power cords use the international IEC 60320 connector family. Two pairs matter for Server+:

• C13 and C14. The C14 is the three-pin inlet on the equipment; the C13 is the socket on the cord that plugs into it. It is the familiar connector on most computer and monitor power cords, rated at 10 amps under IEC standards (15 amps under North American ratings). Most 1U and 2U servers use it.
• C19 and C20. Larger, with rectangular pins, and rated at 16 amps under IEC (20 amps in North America). They are used for high-power equipment such as large servers, blade enclosures and some storage systems, and often for the connection between a PDU and its supply.

Rack PDUs usually provide C13 and C19 outlets, so the cords for rack equipment are C14-to-C13 or C20-to-C19 jumpers rather than cords with a wall plug. Using the right cord matters: forcing a high-power device onto an undersized connector or cord is a fire risk, not merely a fit problem.

Adding up the load in watts and VA, and leaving headroom for a failure

Two units describe power, and UPS units are rated in both.

• Volt-amperes (VA) is apparent power: the voltage multiplied by the current.
• Watts (W) is real power: the power actually used to do work.

The two are related by the power factor, which is at most 1: watts equal VA multiplied by the power factor. Modern server power supplies have a power factor close to 1, so their watts and VA are close. A UPS might be rated at 3,000 VA and 2,700 W, and the equipment on it must stay within both figures.

To size a UPS or a circuit, add up the real load of everything connected to it. Use measured figures where you can, from a metered PDU or the server's management controller, not the rating printed on the power supply: a server fitted with two 800 W supplies might draw 250 W in normal use. The supply's rating is its maximum capacity, not its consumption.

Then leave headroom. Keep a UPS well below its rated capacity, both to extend battery runtime and to leave room for growth, and remember the A/B rule from the first section: when one feed fails, the other must carry everything. A design that only works while every component is healthy has no redundancy at all.