The data centre power chain, from utility to rack

Power reaches a server through a chain: the utility service and switchgear, a transfer switch that can swap to a generator, a UPS that bridges the gap, then distribution units, rack power strips and finally the server's own power supply. Each link has a job, a way to fail and a way to be maintained, and the reliability of the room is set by the weakest one.

The chain at a glance

Every data centre, from a single server room to a large hall, uses some version of the same sequence. Larger sites duplicate parts of it for redundancy; smaller ones combine stages, for example a UPS with built-in distribution. The diagram shows the order that power passes through.

Utility service and switchgear

The chain starts where the utility's conductors meet the building. Larger sites often take service at medium voltage and step it down on site; smaller ones take low voltage directly. The main switchgear holds the service disconnect, metering and the breakers that feed the rest of the building. Some sites bring in two utility feeds, ideally from different substations, though true independence depends on the utility's network and is worth asking about directly.

The switchgear also feeds loads that are not on the UPS: lighting, general receptacles, and often the cooling plant. How cooling is fed matters. If the chillers are on utility and generator only, the room keeps computing during the seconds before the generator takes over, but cooling stops. In a dense room, temperatures rise quickly, so some designs put fans, pumps or controls on UPS as well.

Generator and transfer switch

The generator is the long-duration backup. When the utility fails, the generator starts, comes up to speed and voltage, and the automatic transfer switch moves the load to it. When the utility returns and stays stable, the switch transfers back and the generator runs on to cool down. Larger sites parallel several generators on a common bus. Our guide on transfer sequences walks through the timing in detail.

The generator cannot pick up load instantly. The gap between the utility failing and the generator carrying the load is what the UPS exists to bridge.

The UPS

The UPS sits between the transfer switch and the IT load. In the double-conversion design used in most data centres, incoming AC is converted to DC, which also charges the batteries, and an inverter converts it back to clean AC for the load. If the input fails, the inverter keeps running from the batteries with no break. A static bypass lets the load run on raw input power if the UPS has an internal fault, and a separate maintenance bypass lets the UPS be isolated for service. See UPS topologies for how the designs compare.

Battery runtime is usually sized to cover the generator's start and transfer with margin, plus enough for a second attempt or an orderly shutdown if the generator fails. Longer runtimes mean more batteries, more space, more weight on the floor and more ventilation or cooling for the battery room.

Distribution in the white space

From the UPS, power goes to power distribution units (PDUs). In North America a floor-standing PDU usually contains a transformer that steps 600 V or 480 V down to 208/120 V, plus panelboards with branch breakers. Remote power panels (RPPs) extend those panelboards closer to the racks. Overhead busway is an alternative: a powered track along each row with plug-in tap-off boxes, which makes adding or moving circuits quicker.

Keeping the UPS and batteries out of the white space keeps their heat, maintenance traffic and battery hazards away from the IT equipment. Keeping the PDUs close to the racks keeps branch circuits short and makes it easy to see which breaker feeds which rack.

Rack PDUs and the server power supply

Inside each rack, a rack PDU (a power strip built for data centre use) takes one branch circuit and spreads it across many outlets. Metered and monitored models report current, so operators can see how close each circuit is to its limit. In a dual-path design, each rack has two rack PDUs, one per path. Our guide on rack PDUs and power monitoring covers the choices.

The last link is the server's power supply, which converts AC to the low DC voltages the electronics use. Dual-corded servers have two supplies, each able to carry the server alone, and share load between them.

Where energy is lost

Every conversion loses a little energy as heat: transformers, the UPS's rectifier and inverter, the server power supplies. Those losses are paid for twice, once as electricity and again as cooling to remove the heat. Power usage effectiveness (PUE), the ratio of total facility energy to IT energy, is the common way to track it. Practical ways to reduce the losses include right-sizing UPS capacity so it is not running at a very light load, using efficient transformers, distributing at higher voltages where the equipment allows, and keeping cooling matched to the actual load.

Questions to ask about your own chain

• Is there an up-to-date single-line diagram that matches what is installed?
• Which loads are on UPS, which on generator only, and which on utility only? Is cooling where it needs to be?
• How long will the batteries carry today's load, and when were they last tested?
• When was the generator last tested under load, and does the transfer switch get exercised?
• Can the UPS be bypassed for maintenance without dropping load, and does staff know the procedure?
• Is each PDU, RPP and rack circuit labelled so the source of every rack can be traced?
• How close is each circuit and each UPS to its limit, and who watches that number?

How ehZee can help

We review existing power chains for capacity, single points of failure and maintainability, and design new or expanded systems from the service to the rack. Drawings for permit are sealed by a licensed professional engineer where required, and installation is done by licensed electrical contractors under permit. See our data centre electrical design service.

Common questions