What Is a Data Center PDU? Power Distribution Guide

A data center PDU distributes electrical power from a UPS, generator-backed circuit, or other upstream source to servers and network equipment. Basic models provide outlets; metered, monitored, and switched models add readings, alerts, and remote control. A PDU distributes power, but it does not replace the UPS that keeps power available during an outage.

The rack’s power traffic controller

I used to think of a PDU as a long power strip with better mounting hardware. That description works until you need to answer a simple question during an incident: which device is drawing power, through which path, and what else will disappear if you switch it off?

A PDU, or Power Distribution Unit, takes power from an upstream source such as a UPS, distribution panel, or generator-backed circuit and distributes it to servers, switches, storage systems, and other rack equipment. Depending on the model, it can also measure current, voltage, active power, energy consumption, environmental values, and outlet state.

Basic models only provide outlets. Metered models show readings locally, while monitored models expose measurements to a management system. Switched models can turn individual outlets on and off remotely.

A PDU does not replace a UPS. The UPS keeps supplying power for a limited time when the incoming feed fails; the PDU distributes that available power. Different jobs.

PDU and UPS: two different points in the power chain

A UPS uses batteries or another energy-storage system to maintain its output when the input supply fails. A PDU takes that output and distributes it to several devices, sometimes with measurement and remote control.

You can run a few devices directly from a UPS, but a rack becomes harder to manage as the number of power supplies grows. You need enough outlets, sensible cable routing, and a clear way to identify each connection.

A common power chain looks like this:

utility or generator → UPS → distribution panel → PDU → device power supply

Larger facilities may have dual electrical paths, separate UPS groups, and multiple distribution panels. The diagram matters more than the labels on the equipment. If both power supplies in a server connect to the same PDU, you have two cables but not meaningful power-path redundancy.

Draw the power path – Show every UPS, panel, PDU, circuit, and cable between the source and the server power supply.

The PDU types you are likely to meet

PDU selection is not just outlet counting. I look at the input connection, phase arrangement, current rating, measurement needs, outlet types, and management requirements together.

Basic PDUs keep the job small

A basic PDU supplies power through fixed outlets and normally has no network connection, current measurement, or remote switching. It can be perfectly reasonable for a small rack whose load is already understood and monitored elsewhere.

The simpler design also means fewer management components to configure. That is useful. You still need to monitor the rack through the UPS, panel, or another measurement point, and you cannot remotely confirm which outlet contains a particular server.

Simple does not mean unplanned.

Metered models show local readings

A metered PDU usually displays total current and may display voltage, active power, apparent power, or accumulated energy. Some models provide readings only on a local screen and have no network management.

Energy consumption is commonly expressed in kWh. Instantaneous load is expressed in W or kW. At 230 V and 10 A, the apparent power of a single-phase load is approximately 2.3 kVA. Active power can be lower because it also depends on power factor.

Monitored PDUs send data to your tools

A monitored PDU exposes sensor data over the network. Depending on the product, that may be through SNMP, HTTPS, or a vendor-specific API. I prefer to bring the readings into the same monitoring system as servers and UPS devices, because an isolated display is easy to miss during a busy shift.

SNMPv3 provides authentication and privacy mechanisms described in RFC 3414. Older community-string-only configurations should not be treated as an adequate boundary for a management device.

Keep the management interface off the public internet. Use a management VLAN, access-control rules, strong credentials, and a VPN or restricted administrative network.

Switched PDUs can cut power remotely

A switched PDU can turn individual outlets on and off remotely. That is useful for controlled recovery when a device has stopped responding, but it makes accurate labels and access control essential.

Switching the wrong outlet, stopping several systems that share one PDU, or cutting power while storage is writing can cause real damage. An orderly operating-system shutdown comes first whenever it is possible.

I would not describe a switched PDU as a remote restart button. It is a power-control system.

Label the model correctly – Separate basic, metered, monitored, and switched features in the purchase request; “smart PDU” is not a technical specification.

Single-phase and three-phase choices

Single-phase PDUs are common in small and medium-sized racks. Three-phase models can support higher power density and distribute load across phases, but they bring more installation and balancing work.

A qualified electrician should handle the electrical design and installation. The PDU label alone cannot tell you what is safe in the installed circuit.

Check the input connector, voltage, phase arrangement, circuit breaker, neutral and protective connections where applicable, and the manufacturer’s continuous-load guidance. Do not treat the theoretical maximum as a sensible continuous operating target.

Not every outlet on a three-phase PDU necessarily uses the same phase. Record the load per phase as well as the total. A comfortable total can hide an overloaded phase.

What to do – Record total load and per-phase load separately, then review both whenever you add or move equipment.

How much PDU capacity does a rack need?

Start with measured operating consumption when you can. The “800 W” label on a server power supply normally describes the supply’s capacity, not the server’s constant draw.

For a basic single-phase estimate:

S (VA) = V (volts) × I (amps)
P (W) = V × I × PF

PF is the power factor. At 230 V, 16 A, and a power factor of 0.9, the theoretical apparent power is 3,680 VA and the approximate active power is 3,312 W.

That number is not permission to fill the circuit to 3,680 VA continuously. The usable limit comes from the circuit protection, PDU rating, upstream UPS, connector, cable, equipment documentation, and applicable electrical requirements.

Include these values in the capacity plan:

• Measured operating consumption from servers, switches, and storage systems.
• Short increases during startup or disk-intensive workloads.
• The distribution of A and B feeds for equipment with two power supplies.
• Capacity reserved for future devices.
• The current limits of the PDU, UPS, outlets, circuit, and cable connectors.

If you lack measurement data, document the worst-case assumption and verify actual consumption after installation. A correct calculation does not make a single-PDU rack redundant.

What to do – Compare estimated wattage with PDU or UPS readings after installation and update the capacity record with measured values.

A and B feeds only count when the paths are independent

With a dual-power-supply server, connect one cable to the A PDU and the other to the B PDU. For meaningful redundancy, the two PDUs should also trace back to separate circuits, distribution panels, and UPS paths where the facility design supports it.

Two different PDU names do not prove independence. Both units may still share one wall outlet, circuit breaker, panel, UPS, or upstream generator path.

In my home lab, I keep the power labels visible on the side of the rack rather than hiding them behind the cables. During a short outage in İzmir, I did not have to guess which devices were on the small UPS and which were not. That is a modest arrangement, but the habit scales well.

An ATS can provide two inputs for equipment with only one power supply. It selects an available source and passes it to the single-feed device. An ATS can fail too, and it may not tolerate every device’s inrush current, so check transfer time, input range, rated current, and protection requirements.

Check the load-sharing mode of dual-PSU servers as well. Some supplies share the load evenly; others keep one supply at a low load until the other fails. If the A and B readings do not match the expected behaviour, inspect the cabling and the server’s firmware or power settings.

What to do – During planned maintenance, disable the A and B paths one at a time and verify that every redundant server continues running.

Keep PDU management away from the public internet

Treat a managed PDU like any other network appliance. Change the default password during initial setup, place its management address on a restricted VLAN, and permit access only from the administrative systems that need it.

Before a firmware update, record the installed version, back up the configuration, and document the rollback method. Confirm what happens to the outlets while the management card restarts. Some products keep power distribution active; behaviour varies by model.

For SNMP monitoring, request only the values you need. Useful alerts may cover total current, per-phase current, voltage, temperature, humidity, and outlet state. “Device unreachable” is not enough by itself. A rising temperature or phase imbalance may warn you before an outage.

Keep management-plane events in an audit record. Failed logins, outlet switching, firmware changes, and user changes should have timestamps from a reliable NTP source.

Security here is not complicated. It is easy to neglect.

Outlet types, cables, and rack placement

Outlet types such as IEC 60320 C13 and C19 must match the equipment power cables. C13 connections are common on lower- and medium-power equipment, while C19 connections are used for devices requiring higher current. Check the PDU input connector with the same care.

A vertical 0U model can mount beside rack equipment without using rack units. A horizontal 1U or 2U model consumes rack space, but cable access may be easier in some cabinets.

Account for cable weight, bend radius, and airflow. Power and data cables cannot always be separated completely, but tidy routing, suitable cable lengths, and clear labels reduce maintenance mistakes.

A cable with no label on either end becomes a small research project during an incident. Label the rack, device, PSU path, PDU, and outlet number. Use material that tolerates heat and does not fade easily.

What to do – Put the device name, PSU path, PDU name, and outlet number on both ends of every power connection.

Monitoring and maintaining a PDU

Maintenance is more than checking whether the equipment is still running. At monthly intervals, or according to your risk policy, review total load, phase balance, alarm history, temperature, humidity, connections, and visible signs of cable heating.

Physical inspection must be performed by authorized personnel under safe, de-energized conditions where required. Do not turn an electrical inspection into an improvised midnight experiment.

Set warning and critical thresholds separately, and assign an owner to every alarm. I have seen perfectly good monitoring become background noise because nobody knew who was expected to respond.

Keep a record for every power change. When adding a server, record its phase, A/B path, total rack load, and remaining UPS capacity rather than merely choosing an empty outlet.

In my own Grafana dashboard, I keep instantaneous watts, accumulated kWh, current, and alarm history as separate panels. They answer different questions. A single large number looks tidy but hides useful context.

What to do – Trigger each PDU alarm in a controlled test and prove that it reaches both the monitoring system and the person responsible for responding.

Errors I try not to repeat

The most common mistake is choosing a PDU by outlet count alone. A twelve-outlet model is not automatically better than a four-outlet model. Input type, current capacity, measurement quality, and management features may matter much more.

Another mistake is treating PDU and UPS capacity as unrelated. A PDU may be rated for 7 kW while the upstream UPS is limited to 5 kW. The UPS is then the constraint, and the circuit may impose another one.

Remote outlet control is another place where optimism gets expensive. Storage systems, databases, and virtualization nodes need an operating-system shutdown plan before anyone reaches for the power control.

I once labelled a small UPS installation only on the PDU side and assumed I would remember the other end of each cable. I did not. Adding labels to both ends took ten minutes; tracing an unlabeled cable later took much longer.

Power is only one part of capacity planning. Cooling, rack space, server density, and operating cost belong in the same conversation. Before approving a PDU, I want the upstream limit, circuit protection, phase load, outlet compatibility, monitoring method, and failure procedure written down together.

Before Installing a Rack PDU, Check These Items

• Confirm the PDU input connector, voltage, phase arrangement, and circuit rating.
• Measure or estimate the real operating load of every connected device.
• Leave documented capacity for startup peaks and future equipment.
• Verify that the PDU, UPS, circuit, outlet, and cable limits agree.
• Design A and B feeds through genuinely independent power paths.
• Change default credentials and isolate the management interface.
• Label both ends of every power cable and test alarms and failover procedures.

Frequently Asked Questions

Sources

• RFC 3414 – User-based Security Model for SNMPv3 — rfc-editor.org