Data center EPMS and power metering field guide

What an EPMS is and why a data center needs one

An electrical power monitoring system, the EPMS, is the network of power and energy meters and the software behind them that lets a data center see its whole power chain at once. Meters sit at the utility entrance, the main switchgear, the generators, the UPS input and output, and the PDUs, and they report voltage, current, power, energy, and power quality back to a central server. The operator watches the plant on that screen, and trends it over time.

The reason a data center cannot run without it is that the power chain is the building. A commercial office can lose power and send people home. A data hall that loses the critical bus has failed at the one job it exists to do, and the EPMS is how the operations team sees the plant before it fails instead of after. It answers four questions the floor lives on: how much capacity is left, who is using the energy, what the PUE is, and where the fault was when something tripped.

The EPMS is also the memory of the plant. When a breaker trips at 3 a.m., the trend and the event capture are what tell you whether it was a real fault, a load step, or a sag that rode in from the utility. Without the metering history, every event is a guess, and the operations team is reconstructing what happened from alarms and phone calls instead of reading it off the record.

What is the difference between an EPMS and a BMS?

The EPMS is the electrical-specific monitoring system; the BMS is the building management system that runs the mechanical and life-safety side. The BMS watches the cooling, the air handlers, the humidity, the leak detection, the fire and security points, and it does basic facility metering. The EPMS watches the electrical power chain in depth, at a speed and resolution the BMS was never built for.

The split exists because the two jobs have different clocks. A BMS polling a temperature sensor every few seconds is fine, because a chilled-water loop moves slowly. An electrical event moves in cycles, a fraction of a 60 Hz second, so the EPMS meters sample fast, capture waveforms on an event, and log power quality the BMS cannot see. Putting electrical monitoring on the BMS alone gives you the energy number and misses the sag, the transient, and the harmonic that actually explain a trip.

On most data centers the two are separate systems that share data. The EPMS owns the electrical depth and the power quality; the BMS owns the mechanical plant and rolls up a facility view. They integrate so the operations team has one picture, but the depth lives on the electrical side. The data center BMS and controls are their own topic and their own guide; here the focus stays on the metering and the power chain the EPMS owns.

The meters and the metering hierarchy

Metering in a data center is a hierarchy, not a single point, because the power chain is a series of stages and each one earns a meter. The utility entrance gets metered so you know what comes in and what the utility bills. The main switchgear gets metered at the bus and often at each major feeder. The generators get metered so you see what they produce when they run. The UPS gets metered on the input and the output, because the difference is the UPS loss and the bypass picture. The PDUs get metered at the input and, on a good design, at every branch.

The point of the hierarchy is that the numbers have to reconcile down the chain. What the utility meter reads should account for what the switchgear reads, which should account for the UPS and the mechanical load, down to what the PDUs and branches read. When the levels do not add up, you have either a metering error or an unmetered load, and on a data center an unmetered load is capacity nobody is tracking.

Where the meters land is a design decision driven by what the owner needs to see. A colocation provider billing tenants meters at the boundary of each tenant's power. An enterprise tracking PUE meters at the points the PUE calculation needs. A plant chasing fault diagnosis meters deep at the switchgear and the UPS. Read the metering one-line and the points list before you commission, because the hierarchy the design drew is the hierarchy you have to prove reconciles.

What is revenue-grade metering, and when do you need it?

Revenue-grade metering is metering accurate enough to bill money against, and the line that defines it is an accuracy class. The accuracy classes commonly cited come from ANSI C12.20, the standard for electricity meters, whose content has recently been consolidated into ANSI C12.1. It defines classes 0.1, 0.2, and 0.5, meaning the meter reads within plus or minus 0.1, 0.2, or 0.5 percent of the true value at full load and unity power factor. Revenue-grade work commonly calls for Class 0.2 or Class 0.5; confirm the class the project and the local utility or tenant agreement actually require.

Check metering is everything else: meters accurate enough to manage capacity, balance phases, and watch the plant, but not held to the billing class. A typical panel or PDU meter for operations might be a Class 1 device, accurate to about 1 percent, which is fine for seeing load and trending capacity and useless for splitting a power bill between tenants. The two grades serve two different jobs, and the cost difference is real, so you do not pay for revenue grade where check metering does the work.

Where you need revenue grade is anywhere money changes hands on the reading. A colocation provider billing a tenant by the kWh needs revenue-grade meters at the tenant boundary, or the bill does not hold up when the tenant disputes it. An enterprise that only manages its own capacity and reports its own PUE can run check metering at most points and save the revenue-grade class for the utility tie. The mistake is putting a 1 percent meter on a billing point and discovering it when a tenant audits the invoice.

The CTs and PTs: ratio, polarity, and burden

A meter does not measure thousands of amps directly. It reads scaled-down signals from current transformers and potential transformers, and almost every metering error traces back to these. The current transformer, the CT, clamps or bars around the conductor and produces a small secondary current proportional to the primary, at a fixed ratio like 1000 to 5 or 1000 to 1. The potential transformer, the PT or VT, steps the voltage down on medium-voltage systems so the meter sees a safe signal. The meter multiplies the secondary reading back up by the ratio you program into it.

The ratio has to be programmed to match the CT actually installed. A 1000 to 5 CT on a meter set for 800 to 5 reads 25 percent off, and it reads that way forever, looking perfectly stable and confident while it lies. This is the single most common metering defect at commissioning, and it is invisible unless you check the reading against a known load. The programmed ratio and the nameplate ratio on the CT have to agree, every meter, before you trust a number.

Polarity is the other classic install error. A CT has a marked primary side, the H1 or P1 dot, that has to face the source, and the secondary leads, X1 and X2 or S1 and S2, have to land on the meter the right way around. Get it backward and the meter reads power flowing the wrong direction, showing negative kW on a load that is plainly consuming, or it corrupts the power-factor and the per-phase numbers in ways that look like a real electrical problem. The fix is in the wiring, but you only find it if you look.

Burden is the part people forget. A CT is rated to drive a certain load on its secondary, the burden, set by the meter and the wiring resistance. Run the CT leads too far on too small a wire and the burden exceeds the rating, the CT cannot push its full signal, and the accuracy class you paid for is gone. On long CT runs, size the secondary wire for the burden, or move the meter closer.

Is it dangerous to open a CT secondary?

Yes, and it is one of the most dangerous things you can do in a metering installation. A current transformer with primary current flowing and its secondary left open does not just stop working. The core drives hard into saturation on every half cycle, and the open secondary develops a very high voltage across the open terminals, hundreds of volts on a distribution CT and well over a thousand on high-ratio units. That voltage can break down insulation, arc across the open terminals, cook the CT, and injure or kill whoever opened it.

The rule is absolute. You never open an energized CT secondary. Before you disconnect any CT secondary wiring while the primary conductor is carrying current, you short the secondary terminals, with the shorting block the metering CTs are wired through or with a rated shorting jumper, so the current has a path. A shorted CT secondary is safe and reads zero. An open one is a hazard building voltage with every cycle.

This is why metering CTs land on a shorting test block, and why the block exists. When a tech pulls a meter or works on the secondary wiring, the block gets shorted first. The rookie who unbolts a CT lead on a live feeder to swap a meter, without shorting it, is the story every commissioning agent has heard and nobody wants to be. Short it, then open it, never the other way.

What an EPMS meter actually measures

A power meter on an EPMS measures far more than the kWh a utility meter reads. It captures the instantaneous and the accumulated, the real and the apparent, and the quality of the power, all from the same CT and PT signals. The core set is voltage and current per phase, real power in kW, apparent power in kVA, reactive power in kVAR, power factor, energy in kWh, frequency, and the harmonic distortion. Demand, the averaged peak over a window, comes off the same meter and drives the demand side of a utility bill.

The relationships are worth carrying in your head, because they explain what the numbers mean. Real power, the kW, is the work the load actually does. Apparent power, the kVA, is the product of volts and amps the conductors and gear have to carry. Reactive power, the kVAR, is the part that swings in and out of inductive and capacitive loads without doing work. Power factor is the ratio of real to apparent, and on a data center full of switch-mode supplies it tells you how much of your conductor capacity is doing useful work versus carrying harmonic and reactive current.

Total harmonic distortion, the THD, is the measurement a data center cares about more than most facilities. The IT load is nonlinear and pushes harmonic current back upstream, and the meter that logs THD on voltage and current is how you catch a harmonic problem before it overheats a transformer or a neutral. The PDU and RPP guide covers why those harmonics matter at the transformer; the EPMS is where you watch them across the whole chain.

kW

Real power, the actual work the load does

kVA

Apparent power, the volts times amps the gear must carry

kVAR

Reactive power, the part that swings without doing work

THD

Total harmonic distortion, the harmonic content on voltage or current as a percentage

Branch circuit monitoring at the PDU and RPP

Branch circuit monitoring, BCM, is the bottom of the metering hierarchy, where the EPMS sees power per circuit instead of per panel. A current transformer on every branch breaker at the PDU or RPP reads the current that circuit draws, and the metering board reports amps, power, and energy per circuit up to the EPMS and the DCIM. It is how a data center watches load and phase balance one rack at a time, which is the resolution capacity management actually needs.

The commissioning trap with BCM is the mapping, and it is the same trap whether the data lands on the EPMS or the DCIM. Each CT has to be tied to the right breaker, and each breaker to the right rack or tenant, or the screen shows confident, precise, wrong numbers for the life of the building. The PDU and RPP commissioning guide covers the branch monitoring mapping in detail, because that is where the per-circuit metering is set up and proven against a known load.

From the EPMS side, branch monitoring is what turns the metering hierarchy from a handful of points into a full picture. The utility, switchgear, UPS, and PDU meters tell you the plant is healthy. The branch meters tell you which rack is near its breaker, which phase is loading up, and where the next watt of capacity actually is. Without it, capacity management stops at the PDU and the rack-level decisions get made by walking the floor with a clamp meter.

What does power quality monitoring capture?

Power quality monitoring captures the disturbances that a plain energy meter never sees: the sags, the swells, the transients, and the harmonics that ride in on the power and explain why something tripped. The reference for how these are defined and measured is IEEE 1159, the recommended practice for monitoring electric power quality, which classifies the phenomena by their shape and duration. A voltage sag is a drop to between 0.1 and 0.9 per unit for half a cycle up to a minute. A swell is the opposite. Transients are the fast spikes, and harmonics are the steady distortion the IT load creates.

The piece that makes power quality monitoring worth its cost is the waveform capture on an event. A power quality meter watches the voltage and current continuously, and when a disturbance crosses a threshold it triggers and stores the actual waveform, cycle by cycle, around the event. That oscillographic capture is the difference between knowing a breaker tripped and knowing why. You can see the sag that dipped the bus, the inrush that followed, or the transient that came down the utility, with a timestamp, instead of guessing.

On a data center this is fault diagnosis, not a nicety. When a UPS transfers or a rack drops, the question is always whether the fault was inside the building or rode in from the grid, and the waveform capture answers it. A sag that shows up at the service and ripples down the chain came from the utility. A transient that appears only downstream of a piece of gear started in the building. The meter that captured the waveform turns a finger-pointing meeting into a five-minute read of the event record.

How is PUE measured, and where do the meters go?

PUE, power usage effectiveness, is total facility energy divided by IT equipment energy, and the number is only as good as the two meters behind it. A PUE of 1.0 would mean every watt into the building reaches the IT load; real data centers run above that because cooling, lighting, and losses consume the difference. The metric comes from The Green Grid, and what matters for metering is that it defines where the two measurements are taken and how often.

The Green Grid sets three levels of PUE measurement, and they differ in where the IT load is metered and how continuously. At the basic level, IT power is read at the UPS output and total facility power at the utility input, sampled as little as monthly. At the intermediate level, IT power moves down to the PDU and the sampling tightens to daily. At the advanced level, IT power is metered at the server, total facility is the data-center input less shared loads, and the EPMS samples it continuously. Every level totals the facility at the utility input; what changes is the IT measurement point and the frequency.

The practical lesson for commissioning is that PUE is a metering claim, so the meters that feed it have to be the right accuracy and the right points, and they have to reconcile. A PUE quoted off a UPS-output reading is a different number than one off server-level metering, and comparing them is comparing two different measurements. Pin down which level the project is reporting to, confirm the meters that feed both the numerator and the denominator, and prove they add up before anyone publishes a PUE.

The meter network, the protocols, and the EPMS server

The meters are only useful if they can talk, and they talk over a handful of standard protocols the EPMS server collects. The common ones on a data center are Modbus, in its RTU serial and TCP/IP network forms, and BACnet, in IP and MS/TP, which is the protocol the BMS side speaks. On medium-voltage gear and protective relays you also see IEC 61850, the substation automation standard built for intelligent electronic devices. Some intelligent PDUs and branch monitors report over SNMP instead. The point is to confirm every meter speaks a protocol the EPMS can read before you count on the point.

The physical network matters as much as the protocol. Serial Modbus meters daisy-chain on an RS-485 loop with address, baud, and termination that all have to be set right, and one misaddressed meter or a missing terminating resistor can take down a whole loop. Networked meters land on the monitoring network with IP addresses that have to be managed. Either way, the comms are a commissioning task in their own right: prove every meter is reachable, addressed correctly, and reporting the points the server expects.

The EPMS server is where it all lands. It polls the meters, stores the trends, drives the alarms, and presents the screens the operations team runs the plant on. On larger or mission-critical plants this is a SCADA-grade system with redundancy, because the monitoring itself becomes critical infrastructure once the operators depend on it. A blind operator on a healthy plant is one bad event away from a blind operator on a failing plant, so the server and its network get the same redundancy discipline as the gear they watch.

Alarming across the power chain

The trend is for diagnosis; the alarm is for the moment it matters. An EPMS earns its keep when it tells the operations team that a source has dropped, a breaker has changed state, or a load has crossed a threshold, before the consequence reaches the floor. The alarms ride on the same meters and on status inputs from the gear: a utility loss, a generator start, a transfer switch changing position, a breaker tripping, a phase loading past its limit.

Status points are half of good alarming and the half that gets skipped. A meter reading current is one thing; a contact telling the EPMS whether a breaker is open or closed, or which source an automatic transfer switch is on, is what lets the operator see the topology change in real time. When the utility drops and the ATS swings to generator, the operator should see the source loss, the genset start, and the transfer on the screen as it happens, not infer it from the load disappearing and coming back.

The discipline that makes alarming useful is threshold setting that respects the gear. Warning and critical thresholds set against the breaker rating and the design load give the team room to act before a trip, not a notification after it. Alarms set too tight cry wolf until nobody reads them, and alarms set too loose fire after the damage. Monitoring with no thresholds is a log nobody watches, and a log nobody watches did not need a meter.

How is an EPMS commissioned?

Commissioning an EPMS is mostly proving that what the screen says matches what the copper is doing. The work runs in stages: configure each meter, verify the CT and PT ratios and polarity, prove the meter reads true against a reference, confirm the comms to the server, set and test the alarms, and confirm the trends are logging. Skip any of them and you turn over a system that looks finished and reports fiction.

The heart of it is the point-to-point verification, meter by meter. You confirm the CT ratio programmed in the meter matches the CT installed, and the PT ratio likewise. You check polarity by reading power on a known load and confirming it reads positive and forward, with power factor near unity on a resistive load and no phase reading inverted. Then you confirm the meter reads true, comparing it against a calibrated reference instrument or a known applied load, because a meter that is configured right can still read wrong, and the only way to know is to check it against something you trust.

Then you prove the path end to end. Confirm each meter is reachable on its protocol and address, that the right point on the meter maps to the right tag on the EPMS server, and that the value on the screen matches the value at the meter. Energize a known load and watch the correct channel move, the same way branch monitoring gets proven, so you know the data is tied to the steel. Set the alarm thresholds against the gear ratings, trip a test alarm to confirm it annunciates, and confirm the historical trend is actually logging and retaining data. A meter that reads right but never logged is a meter with no memory when you need the history.

The deficiency that defines a bad EPMS turnover is the meter that was configured but never verified against a reference. It reads, it reports, it trends, and it is off by a CT ratio or a polarity nobody caught, so the capacity numbers, the PUE, and the billing all inherit the error. Verify to a reference, or you have not commissioned the meter, you have only turned it on.

Integration with breakers and protective relays

The EPMS does not stop at meters. It pulls in breaker status and the data from the protective relays, so the operator sees not just how much power is flowing but what the protection is doing. Modern protective relays and electronic trip units are metering devices in their own right, reporting current, voltage, and trip and status information over IEC 61850 or Modbus, and the EPMS reads that alongside the dedicated meters. A breaker that trips reports it, with the cause, and the operator sees it on the same screen as the load.

This is where the monitoring ties back to the protection the power QA work proves. The coordination and arc-flash studies set the relay and trip-unit settings, and those settings are what the EPMS reports against when a fault clears. The power QA pillar guide covers the coordination study and the as-left settings; the EPMS is how the operations team watches that protection live and gets the trip data when a device operates. The two belong together: the study sets the protection, the EPMS watches it.

What the integration buys at a fault is speed of diagnosis. When a relay trips, the operator wants to know which device, on what pickup, with what fault current, and whether it cleared selectively or took more of the bus than it should have. A relay that reports into the EPMS hands that over in the event record. A relay that only sits in the gear means someone has to walk to the switchgear and read the target by flashlight while the floor waits.

Using the data for capacity management

Capacity management is the day-to-day payoff of the EPMS, and it comes down to one distinction: the capacity the plant has versus the capacity it can actually reach. A PDU rated for a load that is loaded unevenly across its phases, or a panel full of breakers that is lightly loaded, is stranded capacity, power the owner paid for and cannot deploy. The metering at the branch and the PDU is what makes that gap visible instead of a surprise when a phase trips.

The number the operations team manages is headroom, per phase, per branch, per PDU, and up the chain. The EPMS trend shows where the load is climbing toward a limit and where it is sitting idle, so the next rack goes where there is real room rather than onto the leg that is already loaded. Without the data, capacity gets managed by spreadsheet and assumption, and the assumption is always more optimistic than the meter. The plant looks half full on paper and trips a phase on the floor.

Stranded capacity is quiet money. A data center that buys a new PDU because the old one looks full, when the old one was just out of balance, has spent capital to fix a metering and balance problem. The EPMS turns capacity from a guess into a reading, and the discipline is to use it: track the per-phase loading, balance as the floor fills, and let the meter, not the breaker trip, tell you where the limit is.

The as-built, the turnover, and the maintenance the owner inherits

The EPMS turnover is a meter list, not a vague handover. The owner inherits a system they have to trust, so the record has to let them reconstruct it: every meter, where it sits, the CT and PT ratios it is programmed for, its accuracy class, the protocol and address it speaks on, and the proof that it was verified against a reference. Tie that list to the electrical one-line so each meter on paper maps to a point on the chain, because a meter list that does not match the one-line is a list the operations team cannot use.

The as-built matters because the plant changes. CTs get swapped, meters get replaced, PDUs get added as the hall fills, and every one of those changes a ratio or an address the EPMS depends on. If the turnover record is accurate, the next person can confirm a reading is right by checking the meter against the documented ratio. If it is not, every troubleshooting session starts by reverse-engineering what the meter is actually set to.

The maintenance the owner inherits is real and gets neglected because the EPMS keeps reporting whether or not anyone maintains it. Meters need their configuration confirmed after any gear change, their accuracy re-verified on a cadence for billing-grade points, their firmware managed, and their comms watched so a dropped meter gets noticed instead of quietly leaving a hole in the data. A billing meter that drifts out of its class and bills wrong for a year is a maintenance failure wearing a software disguise.

The EPMS as the operator's eyes

Strip away the protocols and the accuracy classes and the EPMS is one thing: the operator's eyes on a plant that cannot be allowed to fail. A data center runs on the premise that the power chain is watched continuously and that a problem is seen before it becomes an outage. The EPMS is that watch. When it is true, the operations team runs the floor on real numbers. When it lies, they run it on confident fiction, which is worse than running it blind, because they trust it.

That is why the metering deserves the same commissioning rigor as the gear it watches. A perfectly built power chain with a misconfigured EPMS hands the owner an operator who cannot see it. Get the meters right, prove them against a reference, map them to the one-line, and turn over a record the operations team can maintain, and the plant the commissioning team built stays visible for its whole life. Get them wrong and the first real event is when everyone learns the eyes were closed.

What to document

The EPMS record is the meter list and the proof behind it, captured so a reviewer can confirm any reading from the paper. Capture each meter and the configuration it depends on, because the configuration is what a future change has to be checked against. The table below is the spine of the metering turnover set, one row per meter.

Common mistakes

• Programming a CT or PT ratio that does not match the transformer actually installed, so the meter reads confidently wrong.
• Installing a CT backward, so power reads negative or the per-phase and power-factor numbers are corrupted.
• Opening an energized CT secondary without shorting it first, building a high voltage that can arc, damage gear, or injure.
• Turning over a meter that was configured but never verified against a calibrated reference or a known load.
• Putting a check-grade meter on a tenant or utility billing point that needs a revenue-grade accuracy class.
• Leaving meters with comms that were never proven, so points drop out and nobody notices the hole in the data.
• Mapping a meter point to the wrong tag on the EPMS server, so the right value lands under the wrong name.
• Running power quality meters with no waveform capture or no event thresholds, so a fault leaves no record to diagnose.

Field checklist

Standards and references

Meter accuracy follows ANSI C12.20, the standard for electricity meters and its 0.1, 0.2, and 0.5 accuracy classes, whose content has recently been consolidated into ANSI C12.1 for electricity metering. Revenue-grade work commonly calls for Class 0.2 or 0.5, but the required class comes from the project specification and the utility or tenant agreement. Confirm the class and the current standard reference rather than carrying a number from memory.

Power quality monitoring references IEEE 1159, the recommended practice for monitoring electric power quality, which classifies sags, swells, transients, harmonics, and interruptions by shape and duration. The communications follow the protocol standards the meters speak: Modbus, BACnet, and IEC 61850 for substation and protective-relay integration, with SNMP common on intelligent PDUs and branch monitors. PUE follows The Green Grid, including its levels of measurement that fix where the IT load and total facility are metered.

The acceptance testing of the CTs, PTs, and metering during commissioning falls under the NETA Acceptance Testing Specifications, ANSI/NETA ATS, alongside the rest of the power QA covered in the power pillar guide, which includes the CT and PT ratio and polarity verification. The installation is built to the NEC, NFPA 70, as adopted and amended by the jurisdiction, and the meter manufacturer's instructions and the project specification control the specific configuration. Edition numbers and section references shift between cycles, so confirm the edition and any local amendments against the project documents before citing a standard on a submittal.

Units, terms, and acronyms

The metering side of a data center carries its own vocabulary, and the same reading can be labeled differently across a meter, an EPMS screen, and a billing report. The terms below are the ones that travel across the metering, the monitoring, and the turnover record.

EPMS

Electrical power monitoring system, the meters and software that watch the electrical power chain

CT / PT (VT)

Current transformer and potential (voltage) transformer, the devices that scale current and voltage down to meter signals

Revenue grade

Metering accurate enough to bill against, commonly ANSI C12.20 Class 0.2 or 0.5

PUE

Power usage effectiveness, total facility energy divided by IT equipment energy

THD

Total harmonic distortion, the harmonic content on voltage or current as a percentage

kW / kVA / kVAR

Real power, apparent power, and reactive power, related by power factor

BCM

Branch circuit monitoring, per-breaker current and power metering reported to the EPMS or DCIM

Modbus / BACnet / IEC 61850

The common protocols meters and relays use to report to the EPMS server

Shorting block

The terminal block that lets a CT secondary be safely shorted before it is opened

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is the difference between a PDU and an RPP in a data center?
• How long does data center electrical commissioning take?
• What is data center commissioning?
• What is PUE?
• What is data center TCO?