Home › Topics › Power & Environment Monitoring › Temperature & Humidity

Temperature and Humidity: Monitoring the Server Room Climate


Every server in your building is a space heater that happens to compute. Nearly all the electricity it draws leaves as heat, all day, every day. The only thing between that heat and your hardware is an air conditioner chosen by whoever built the room. It was sized for whatever the room held at the time, and set to whatever suited whoever was cold last. When the cooling works, nobody thinks about it. When it stops, you are living inside a physics problem with a deadline.


Climate problems arrive quietly. Servers do not complain about heat until they are already throttling or shutting down. The air conditioner does not announce that its compressor died at midnight; the room just warms, degree by degree, while everyone sleeps. By the time a human walks in and asks why it is tropical in here, you have spent hours of damage that you could have spent responding. I have walked into that room. The air meets you at the door. The first thing you do is look at the wall thermostat, which says everything is fine. Temperature monitoring exists to move that discovery from a human nose to an alert with a timestamp, ideally hours earlier.


This guide covers the climate half of protecting a server room. It explains what temperatures actually matter (and why the thermostat's opinion is not one of them). It covers where sensors have to sit to tell the truth, and what really happens after the AC fails. It covers what humidity does at its extremes and how to set thresholds so alerts arrive with time to act. And it covers the playbook to run when the hot day comes. You will also meet the free temperature sensors you already own without knowing it.


In this guide

  • Safe ranges in plain language
  • Why the thermostat lies
  • Sensor placement that tells the truth
  • Thermal runaway: the clock after AC failure
  • Humidity: the two bad endings
  • Alerting with lead time
  • The AC-failure playbook
  • Free sensors you already have

Safe ranges in plain language


First, agree on which number we are even talking about. The temperature that matters to a server is the temperature of the air entering its front — the intake temperature. Everything else in the room is scenery. Equipment makers design around intake air. The widely used comfort band for it sits roughly between 18 and 27 °C (about 64 to 80 °F). Most gear will tolerate excursions beyond that band without immediate drama. The point of the range is not that 28 °C kills servers. It is that margin above the comfortable band is your emergency buffer. Running hot in normal times means running with no buffer on the day the AC quits.


Heat also costs you quietly even when nothing fails outright. Fans spin harder and wear faster. Batteries — including the ones inside your UPS, as the UPS guide explains — age measurably faster in warm rooms. Components live at the hot edge of their tolerances, and intermittent oddities become a little less rare. Cool rooms are not superstition; they are you buying slack in a hundred small places at once.


Stability deserves a mention alongside level. A room that swings widely — cold every night, warm every afternoon, colder again when the AC overshoots — stresses hardware in its own way. That is because parts expand and contract with every cycle and connections feel every flex. If your temperature graph looks like a saw blade, the cooling is short-cycling or fighting the room. In that case, smoothing that out is worth a conversation with whoever maintains it. The graph is what makes the conversation possible: "it cycles eleven degrees every afternoon" lands very differently than "it feels off in there."


Humidity has a comfortable middle for the same reason, just with different failure modes at each end. A common target is to keep relative humidity somewhere in the middle band, avoiding both the parched extreme and the swampy one. Relative humidity is how much moisture the air holds compared to the most it could hold at its current temperature. We will get to what each extreme actually does in its own section. For now the summary is: dry air makes sparks, and wet air makes dew. And both are bad news for electronics.


Why the thermostat lies


Here is the trap almost every small server room falls into: someone checks the wall thermostat, sees a civilized number, and concludes the room is fine. But the thermostat is mounted on a wall, at head height, usually near the door or the AC unit itself. And it measures the blended, well-mixed air at that one spot. Your servers are not breathing that air. They are breathing whatever arrives at their own intakes, and in a room with racks, the difference can be startling.


Air in an equipment room is not uniform, because the equipment itself keeps sorting it. Servers inhale on one side and exhale hot on the other, so the room develops hot and cold regions. That is a miniature version of the hot-aisle and cold-aisle layout big data centers formalize. Hot exhaust rises, so the top of a rack runs warmer than the bottom. Worst of all is recirculation. When hot exhaust finds a path back around or over a rack into an intake, one server ends up drinking another server's exhaust. In that case, a device can run ten degrees above the wall number while the thermostat radiates smug satisfaction. Meanwhile the AC's own sensor often sits in the return air stream or right in its cold blast, giving it the most flattering possible view of its own performance.


The lesson is not that thermostats are useless — it is that the room does not have a temperature. It has a temperature map, and the only points on the map that matter are the intakes. Monitor those.


Tollgate Foods found out why their server room hit 40 °C every August from a graph, not from a person. The building's cooling plant dropped to a holiday schedule for the long weekend at the end of the month. The server room was on the same plant. The wall thermostat by the door read 24 °C the whole time, because the door was where the corridor air came in. The intake sensor at the top of the main rack told a different story. It showed a climb that started on Friday evening and peaked on Sunday afternoon, three years running. Each Monday morning, the graph dipped when the plant came back. Nobody had been in the building to feel it. The fix was one line in the holiday schedule. The diagnosis was the sensor.


Sensor placement that tells the truth


Networked temperature sensors are inexpensive, and placement decides whether they inform you or flatter you. One sensor is an opinion; two that agree are a fact, and two that disagree are a diagnosis. The rules of thumb:


  • Put the primary sensor at a rack intake, high up. The front of a rack, roughly at the top third, is where warm trouble shows first — it is the spot recirculating exhaust reaches earliest. This one sensor is worth more than three on walls.
  • Use more than one. A second sensor at a different rack, or at the opposite end of the room, catches localized hot spots. It tells you whether "the room is hot" or "this corner is hot" — different problems with different fixes.
  • Keep sensors out of the AC's cold blast. A sensor bathed in supply air reports the AC's output temperature, not the room's. The same goes for mounting one directly on warm equipment: you are then measuring the equipment, not the air.
  • Cover the suspected hot spot. Every room has one — the corner behind the tallest rack, the top of the densest one, the alcove with no airflow. If you are unsure where it is, a cheap handheld thermometer on a slow walk around the room will find it. Put a sensor where the walk surprised you.
  • Poll at a calm rhythm and keep the history. Climate moves slowly compared to networks: readings every minute or every few minutes are plenty. What matters more is retention — keep months of it, because the value of climate data compounds. This summer's graph versus the last one is how you notice the AC aging before it fails.
  • Add a reference sensor outside the cooled space if you can — the hallway or adjacent office. When the room alarm fires, this tells you instantly whether the building is hot (AC plant, summer failure) or only your room is (your unit, your problem).

The thread running through all six rules is that a sensor measures whatever is nearest to it. And the air conditioner is very good at being nearest. A probe in its supply stream reports the AC's press release. A probe at an intake reports the room as the servers experience it, and that is the only review that counts.


Thermal runaway: the clock after AC failure


The scenario the whole setup exists for begins with the air conditioner stopping — compressor failure, tripped breaker, building maintenance switching things "just for a minute." What happens next is straightforward and unforgiving. Your equipment keeps converting electricity into heat at exactly the same rate as before. The heat now has nowhere to go. The room is a sealed box with a heater in it that you cannot switch off quickly. Temperature does not wander; it climbs.


How fast is entirely a question of ratios: how much heat the equipment produces versus how much room, air, and wall mass there is to soak it up. A lightly loaded closet with a couple of switches may drift upward lazily for hours. A small room packed with servers can become dangerously hot in well under an hour. The same heat output that needed constant mechanical cooling to hold steady is now compounding in a few cubic meters of air. The early part of the climb is the steepest part of your response window. Two accelerants make the late part worse. As the room heats, every device's fans spin up to move more of the same hot air. Recirculation gets more aggressive too, so intake temperatures climb faster than the room average. Eventually servers protect themselves by throttling and then shutting down — which finally does reduce the heat, in the least useful way imaginable.


You cannot know your room's exact curve in advance. But you can know its shape: monitor intake temperatures continuously for a few weeks and you learn your baseline and your daily rhythm. Then the first AC hiccup — there is always one — hands you a free measurement of how fast your room climbs when cooling pauses. File that number. It is the honest basis for every threshold in the next section, and it is the difference between guessing your response window and knowing it.


One timing detail makes all of this personal: cooling loves to fail when nobody is there. A compressor that dies on Friday evening has the entire weekend to work with. A monitoring gap of two days converts a repairable hiccup into a Monday of cooked equipment. This is why climate alerts must be wired to reach a human after hours. The alert that lands in an inbox nobody reads until morning protects nothing but the timeline in the post-incident writeup.


Humidity: the two bad endings


Humidity fails at both ends, gently but genuinely. At the dry end, static electricity becomes the hazard. Dry air lets charge build up on people, chairs, and packaging. A static spark that a human barely feels is more than enough to damage components. The danger window is mostly while handling hardware: open cases, RAM in hands, boards on desks. Very dry rooms plus lots of hands-on work is the combination to avoid, and grounding straps plus a humidity number you actually watch are the boring, effective answers.


At the wet end, the enemy is condensation. Air can only hold so much moisture, and the ceiling drops as air cools. The temperature where moisture starts falling out as liquid is the dew point. Humid room air meeting a cold surface — chilled water pipes, the AC's own coils and ducts, equipment in a cold aisle — produces droplets exactly where you want them least. High humidity also feeds slower rot: corrosion on contacts and connectors that shows up months later as flaky, maddening intermittent faults. Sustained high readings, or a rising trend after weather changes or a cooling change, deserve investigation rather than a shrug. And if water is a broader worry in your room, the guide on water and door sensors continues that thread.


Alerting with lead time


The purpose of a climate alert is not to announce damage — it is to hand you the problem while it is still cheap. That means thresholds must be staged, with the first stage set well below anything harmful. A sensible ladder looks like this (starting points to tune, not commandments):


StageIntake temperature (example)MeaningResponse
AdvisoryAbove ~27 °C / 81 °FWarmer than your normal; buffer shrinkingLook at the trend today, not right now
WarningAbove ~32 °C / 90 °FSomething is wrong with coolingAct now: start the playbook
CriticalAbove ~38 °C / 100 °FDamage territory approachingShed load, consider shutdowns
Rate-of-riseClimbing several degrees within minutesCooling has likely stopped entirelyTreat as warning immediately, whatever the absolute number

That last row is the underrated one. A rate-of-rise alert triggers on how fast the temperature is climbing rather than the value it has reached. It is the earliest bell you can own, because an AC failure announces itself by slope long before any absolute threshold trips. A room sitting steady at 26 °C is a mild to-do item. A room that went from 21 to 26 in a quarter of an hour is an emergency in progress, even though both rooms currently read 26. If your monitoring system can alert on change over time, use it here; this is the textbook case for it.


Set thresholds per sensor, against that sensor's own normal, rather than one number for the building. A device inlet that always reads a few degrees above the room needs its stages shifted up by the same offset, or it will alert every week about nothing. The sensor in the naturally cool corner needs them shifted down, or it will stay silent while the far rack suffers. Baselines first, thresholds second — the order matters.


Tip: alert on the humidity extremes with generous staging too, but resist tight humidity thresholds. The value drifts with weather and seasons, and a twitchy humidity alert is how climate monitoring earns itself a mute. Wide bands, sustained-for-a-while conditions, and trend review beat precision theater.


The AC-failure playbook


When the warning fires, the minutes go further if nobody has to invent the response. The playbook is short enough to memorize and belongs written down anyway — the guide on first runbooks shows the format. In order:


  • 1. Confirm it is real. Two sensors agreeing, or one sensor plus a device's own inlet reading, beats one possibly-lying probe. Thirty seconds here prevents the 2 a.m. drive for a sensor that fell off its mounting.
  • 2. Open the doors. The building's air is now your heat sink. Propping the room open swaps a sealed hot box for a bigger, cooler one, and buys real time in small rooms. This conflicts with any door-security alert you also have; overriding it on purpose is part of the playbook.
  • 3. Move air. A box fan in the doorway pushing hot air out does honest work. Where you place it matters less than that it exists in the room already, plugged-in-able in seconds, not in a storeroom two buildings away.
  • 4. Shed load. Every watt you turn off is heat that stops arriving. Shut down the expendable tier first — test boxes, spare capacity, anything whose absence hurts nobody. That is exactly the list the graceful shutdown guide had you write. In a bad climb, continue up the tiers; a server shut down cleanly at 35 °C beats one that cooked at 45.
  • 5. Call the fixer. The name and number of whoever repairs your cooling — facilities, the HVAC contractor, the landlord's emergency line — belongs in the runbook, not in someone's phone. This call starts the actual repair; everything above only buys time for it.
  • 6. Watch the slope. Your sensors now tell you whether the measures are working. The climb flattening means you have time. The climb continuing means proceed to deeper load shedding. This is monitoring paying for itself in real time.

Step two will trip the door alarm if you have one, and that is fine. Overriding a control deliberately, with a reason you can say out loud, is exactly what incidents are for. The time to worry is when the door alarm is being overridden on a quiet Wednesday with no incident in sight, which is a different guide.


Free sensors you already have


One more source of truth costs nothing. Most managed switches, routers, and servers already measure their own temperatures and will report them over SNMP alongside everything else you poll. Many report an inlet or intake reading — the air arriving at the device — which is exactly the number this whole guide cares about. Two caveats keep them honest. A sensor inside a chassis reads somewhat above the room because the device warms it. And models differ in where the probe sits. So treat absolute values gently and trends seriously. A switch inlet that has read a steady offset above your room sensor for months is a calibrated instrument in every way that matters. When it climbs, believe it. In rooms with real sensors, device inlets add free coverage and a second opinion. In the places that will never get a dedicated sensor — the remote wiring closets — they are often the only thermometer you have. The guide on the closet nobody checks builds on exactly that.


Where to go next


Measure at the intakes, in more than one place. Stage your thresholds so the first alert arrives with hours of slack. Add a rate-of-rise bell for the day the cooling dies outright. Keep humidity in the comfortable middle, and have the playbook written before you need it. Climate problems will still happen; compressors do not care about your monitoring. But they will happen to a team with lead time, a plan, and a graph that shows whether the plan is working, rather than to a wall thermostat that says everything is fine. The rest of Power & Environment Monitoring covers the neighboring disasters. Those include the UPS that claims more runtime than it has, the automated shutdown that saves your data, and the water creeping toward the floor drain.


How do you monitor data center and server room temperature and environmental conditions, temperature on Linux servers, and an APC Smart UPS?

Three sources, all SNMP or near it. Environmental probes in the room report temperature and humidity over SNMP; switches and servers report their inlet temperature over SNMP too, and servers also over IPMI or Redfish, which covers Linux servers without installing anything on them. An APC Smart UPS with a network card reports load, battery and runtime over SNMP with the APC MIBs. Put all three on the same dashboard and the room's story reads from the intake up.


Put this into practice with SysPeep Monitor

Temperature and humidity sensors, and the inlet readings your switches and servers already take, all speak SNMP. That makes climate a monitoring job like any other. SysPeep Monitor collects those readings on a schedule, graphs the trends, and alerts you at the thresholds you stage. The free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep