Data center equipment monitoring: power and cooling
Data center equipment monitoring starts with the electrical chain, from the utility feed through UPS units and PDUs, and with cooling air measured at server inlets. Keep inlets within 18 to 27°C, the ASHRAE TC 9.9 recommended range. Also track PUE, generator readiness and battery resistance.
Which power chain metrics should a data center monitor?
Log voltage, current, frequency and power factor at the utility feed, UPS input and output, and PDUs. For each UPS, add load percentage, battery state of charge, estimated runtime and battery temperature. DataCenterGuidelines.com lists these as typical points to historize, along with operating mode, bypass and battery events, and communications loss.
Update rates depend on the signal. The same source suggests these starting ranges, which are not mandatory:
- Fast status: 1 to 5 seconds
- Electrical measurements: 5 to 15 seconds
- Energy: 15 to 60 seconds
- Slow diagnostics: 30 to 300 seconds
Treat runtime as an estimate, not a guaranteed autonomy value. Cell-level internal resistance is worth trending too, because a battery can look steady at string level while single cells decline.
Power and cooling signals come from different vendors, so map them to one schema before you correlate events. Our guide on normalizing OEM telemetry shows how.
What cooling parameters should you monitor, and where do sensors go?
Monitor supply and return air temperature, airflow or differential pressure, humidity, and the health of chillers, pumps and CRAH units. HVAC Engineering Canada describes the common practice: control supply air at the cooling unit, then confirm conditions with rack inlet sensors.
Place sensors by purpose:
- Rack inlet temperature: front of the rack at top, middle and bottom, to confirm compliance and reset setpoints.
- Rack exhaust temperature: rear of the rack or hot aisle, to assess delta T and load.
- Differential pressure: across containment or the underfloor, to drive fan speed.
- Humidity or dew point: cold aisle and return air.
- Leak detection: under CRAHs, CDUs, piping and manifolds.
Add vibration and temperature checks on pumps, fans and compressors so mechanical wear shows up early.
What inlet temperature and humidity ranges should you hold?
Hold server inlet air at 18 to 27°C. ASHRAE's thermal guidelines set the recommended moisture range at -9°C dew point up to 15°C dew point and 70% relative humidity. Class A1 equipment has a wider allowable range of 15 to 32°C.
These limits apply to air entering the IT equipment. They do not apply to return air, room average or the cooling unit reading. Operating in the allowable band for short periods is acceptable. Extended operation there can affect reliability, fan energy and server noise.
The upper humidity limit drops to 50% where gaseous contaminants are elevated, because humidity speeds corrosion when sulphur compounds are present.
How do you set alarm thresholds without causing alert fatigue?
Tie thresholds to the ranges above and route each alarm by severity. Use this rule at rack inlets, with three sensors per rack at top, middle and bottom, as Raritan describes:
- Normal: 18 to 27°C.
- Warning: outside the recommended range but inside the allowable range (15 to 32°C for Class A1). Log it for maintenance review.
- Critical: outside the allowable range. Notify someone immediately.
Treat a failed generator start as critical. Treat rising battery resistance as a warning. Do not alert on every small deviation, because noise hides real events.
Three mistakes cause most trouble: alerting everyone on everything, trusting sensors that are flatlined or drifting, and keeping no trend history. Data quality checks and trend rules address all three. Our overview of anomaly detection across mixed equipment covers where those rules break.
Which efficiency and capacity metrics matter?
Track PUE, load per circuit and redundancy headroom. PUE is total facility energy divided by IT equipment energy. The Green Grid says to measure only the data center portion of the utility meter. Office or other building loads make the result faulty.
This matters most when a data center shares a building with offices. Install a submeter on the data center feed and use its reading as total facility energy. Divide that reading by the IT equipment energy. If you use the whole building meter instead, office energy inflates the numerator and overstates PUE.
Watch load per circuit against breaker ratings. Check that, if one path fails, the remaining path can carry the full load.
How should you monitor and test generators and batteries?
Monitor generator fuel level, coolant temperature, oil pressure and starting battery voltage, since each points to a failed-start risk. Low fuel gives time to book a delivery. Falling oil pressure warns of mechanical wear. A weak starting battery is a common cause of failed starts, and the failed-start event itself is a critical alarm.
Test UPS batteries on a schedule. Follow the maintenance and test schedule that the battery maker and your applicable standard set for your battery type, including replacement criteria. Record internal resistance at each test and trend it, because it rises before voltage drops. Keep run hours, events and maintenance records, so you can show each generator was tested and when.
We build the infrastructure layer that turns this telemetry into one machine-readable equipment model, with failure patterns as automatic alerts. See how we unify data center equipment monitoring.
Related resources
SailPlan builds the machine-readable data model that makes every AI tool in your stack actually work. Request a demo to see it in action.