Rack PDUs and power monitoring

A rack PDU is the power strip inside a server rack, and it is the last point where you can measure and control power before it reaches the equipment. Choosing PDUs that report their load, and loading A and B feeds so either one can carry the whole rack, is what keeps a redundant room redundant as it fills up.

What a rack PDU does

A rack PDU takes one branch circuit from a panel, remote power panel or busway and distributes it to the equipment in the rack. It is usually a vertical strip mounted at the back of the rack, with outlets suited to server power cords and a locking plug or a hard-wired connection at the input end. Most include a breaker or fuses that protect groups of outlets.

A rack PDU is not the same as the floor-standing PDU that feeds whole rows. Both are called PDUs, so specifications and drawings should say which one they mean. Our guide to the data centre power chain shows where each one sits.

Types of rack PDU

Rack PDUs are sold in a ladder of features. Each step up adds visibility or control and costs more. The diagram shows the usual steps.

The rack PDU ladder. Each step keeps the features of the one before. Monitored PDUs, which report their load over the network and raise alarms, are where most data centres find the best balance of cost and visibility.

Basic: outlets and overcurrent protection only. Cheap, but nobody knows how loaded it is without a clamp meter.
Metered: a local display shows total current, so a technician at the rack can check the load before plugging in more equipment.
Monitored: the same readings over the network, with thresholds that raise alarms. This is the minimum for any room that is managed remotely.
Switched: each outlet can be turned on or off remotely, useful for rebooting a hung device at an unstaffed site or staggering power-up after an outage.
Outlet metered: measures power at each outlet, so you can see what each device actually draws and charge it back to a tenant or department.

Many PDUs also accept temperature and humidity probes, which turn every rack into a monitoring point for the cooling system.

Sizing the circuit and the PDU

A rack PDU is rated for the circuit that feeds it: voltage, phase, current and plug type. Common choices are single-phase 208 V or 120 V, and three-phase 208 V for denser racks. Three-phase PDUs carry much more power on one cord but spread the load across three phases, so equipment should be plugged in to keep the phases roughly balanced. See three-phase power basics if that is unfamiliar.

Data centre loads run continuously, and continuous loads are commonly limited to 80 percent of the circuit breaker's rating unless the equipment is specifically rated for full continuous load. Check the Canadian Electrical Code as adopted in your province for the rule that applies to your installation.

A worked example

Take a single-phase 208 V, 30 A circuit. Its nominal capacity is 208 × 30 = 6,240 VA. At 80 percent, the usable continuous load is about 5,000 VA. Now give the rack two of these circuits, one from the A side and one from the B side, with all equipment dual-corded. If one side fails, the other must carry the whole rack, so the whole rack must stay under about 5,000 VA. In normal operation, the load splits between the two sides, so each PDU should read no more than about 2,500 VA. A PDU reading 3,500 VA looks fine against its own rating, but it means the rack would trip a breaker the moment the other side is lost.

Choosing the input and the outlets

Decide the input before ordering PDUs, because it ties the rack to a circuit type for years. A locking plug lets a PDU be swapped without an electrician, but the matching receptacle must be installed on the branch circuit, under the busway tap-off or on a cord drop from the RPP. Match the outlet mix to the equipment: most servers use the IEC C13 or C19 style of connector, while some network gear and monitors still use standard household-style plugs. Count outlets for the rack at full build-out, not on day one, and choose a PDU length that fits the rack height. Record each PDU's source panel and circuit number on the PDU label, in the monitoring software and on the panel schedule, so all three agree.

Loading A and B feeds

This is the most important idea on the page, and the one most often missed. In a dual-path rack, the number that matters is not each PDU's load against its own limit. It is the combined load against the limit of one PDU.

Load each side to half, not to full. With both feeds healthy, each rack PDU carries about half the rack. When one feed is lost, the other picks up everything, so each side must normally stay below half of the circuit limit.

Set monitoring thresholds to match. A warning at around 40 percent of the circuit limit and an alarm at 50 percent on each side of a dual-path rack is a common approach; your own thresholds should follow your design and the breaker ratings. The same rule applies further upstream: each UPS, PDU and RPP in a 2N system must stay below half of its capacity. Our guide to N, N+1 and 2N redundancy explains why.

Monitoring that gets used

Monitored PDUs report current, voltage, power in kilowatts, energy in kilowatt-hours and often power factor, per PDU, per phase and sometimes per outlet. That data is only useful if it reaches a system someone watches.

From the rack to a decision. PDUs report over a separate management network to monitoring software. Alarms go to whoever is on call; trends show which racks and circuits are filling up, long before a breaker trips.

Put PDUs on a management network separated from production traffic, change default passwords and keep firmware current. See VLANs and network segmentation.
Name each PDU in the software by rack, side and source panel, so an alarm points straight to a breaker.
Set thresholds per side using the A and B rule, not the PDU's own rating.
Send alarms to a person, not a mailbox nobody reads, and test the path.
Review trends monthly: rising racks, unbalanced phases, sides that drift apart (which can mean a failed server power supply).
Compare total rack PDU load with UPS output; a large gap points to loads you are not measuring.

Common mistakes

Plugging both cords of a server into the same PDU, which quietly removes its redundancy.
Planning capacity from equipment nameplates, which usually overstate real draw. Measure instead, and leave margin for peaks.
Adding a power strip to a PDU outlet to get more outlets. Relocatable power taps plugged into PDUs are generally not acceptable in a data centre and overload circuits nobody is watching.
Leaving outlets unlabelled, so nobody knows what will go dark when a PDU is swapped.
Ignoring phase balance on three-phase PDUs until one phase trips.

How ehZee can help

We specify rack PDUs and the circuits that feed them, set loading rules and monitoring thresholds that match the redundancy design, and review existing rooms where load has grown past what the original design allowed. 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

Do we really need monitored PDUs?

For a room anyone manages remotely, yes. Without them, the first sign of an overloaded circuit is a tripped breaker, and nobody knows how close a dual-path rack is to losing its redundancy.

Is 208 V better than 120 V for racks?

For most servers, 208 V delivers more power per circuit and is slightly more efficient. 120 V remains common for small rooms and some network gear. Check what the equipment accepts and plan circuits accordingly.

Can PDU data be used for billing tenants?

Outlet-metered PDUs can report energy per device or per rack, which supports chargeback. If you bill customers for energy, check whether your province requires approved revenue-grade metering for that purpose.