An overloaded socket, an air conditioner that switches off or a worn-out battery can stop your servers. Practical checks for the power path, UPS choice and upkeep, cooling, water and fire risk, and physical access.
In most businesses the server room is set up once and then left to itself. Yet an overloaded socket, an air conditioner that switches off at the weekend or a battery that quietly reaches the end of its life can stop the servers and every job that depends on them. This article gives you a practical framework for reviewing the physical side of the room: power, uninterruptible power supplies, cooling, water and fire risk, and access.
Power: start by tracing the electrical path
Write down which circuit breaker, which socket and which uninterruptible power supply (UPS) feeds each device in the room. This simple diagram often turns up surprises: the server circuit sharing a breaker with the office heater or kettle, power strips daisy-chained together, or a network switch plugged straight into the wall instead of the UPS.
Where possible, give the server room its own electrical circuit and breaker, and have an electrician check the earthing and the load the circuit can carry. Instead of chaining power strips, use power distribution units (PDUs) fixed inside the rack.
Some servers and storage devices have two separate power supplies. For that redundancy to work, the two cables must not go into the same socket or the same UPS; connect them to different UPSs or different circuits where you can. Otherwise a single fault silences both supplies at once.
Don't forget the network equipment. If the servers keep running on the UPS while the modem, firewall and switches lose power, nobody can reach the systems even though they are up, and the UPS's shutdown signal may never reach the servers.
UPS: the right type, the right size, regular maintenance
Which type of UPS?
UPSs come in three basic designs. They differ in how well they respond to problems on the mains supply and in cost.
| Type | How it works | Typical use |
|---|---|---|
| Standby (offline) | Normally feeds equipment straight from the mains; switches to battery during an outage. | Desktop computers and small devices |
| Line-interactive | Corrects voltage dips and surges to a certain extent; switches to battery during an outage. | Small server rooms and network cabinets |
| Online (double conversion) | Feeds equipment continuously from its own output; mains disturbances never reach the equipment and there is no break when switching to battery. | Sites with unstable power and the most outage-sensitive systems |
The online type gives better protection, but it is usually more expensive and produces more heat while running. That heat needs to go into your cooling calculation too.
Size the UPS in watts
UPS labels show two figures: VA (apparent power) and W (real power). The total draw of the connected equipment must not exceed the UPS's watt rating; looking only at the larger VA figure on the label is misleading. Don't run the UPS at its limit either: the higher the load, the shorter the battery runtime, and there is no room left to add another device. Read the actual load from the UPS display or management interface rather than estimating it.
What should the battery runtime cover?
A UPS is not meant to keep systems running for hours. Its job is to ride through short outages unnoticed and, in a longer one, to give enough time to shut systems down cleanly, or to bridge the gap until a generator takes over if you have one.
For a controlled shutdown, the UPS must notify the servers through its network card or USB connection, and the servers must have shutdown software installed that listens for that signal. Decide the shutdown order: application servers first, then databases, storage devices last; start-up follows the reverse order. Test this sequence for real at least once, at a planned time. If there is a generator, confirm with a test that the UPS works smoothly on the generator's output.
Batteries are consumables
Batteries lose capacity over time, and the UPS front panel does not always show it. Write the battery installation date on the UPS and follow the replacement interval the manufacturer recommends; a high room temperature shortens battery life noticeably. Run the UPS's own battery test regularly and make sure the results arrive at an email address.
A UPS usually does not power the air conditioning. When the mains fails, the servers carry on running on battery while the air conditioner stops, and a small, closed room can heat up quickly. Set the battery runtime and the shutdown timing with this in mind.
Cooling: comfort air conditioning may not be enough
Servers produce heat day and night. A common problem is a room connected to the same central system as the office, which switches off in the evenings or at weekends. Give the room its own air conditioner that can run continuously. Check that its automatic restart after a power cut is enabled; if it isn't, the servers come back on when power returns but the air conditioner stays off.
Measure the temperature at the front of the equipment, where it draws in air, not in the middle of the room. Widely accepted industry guidelines recommend roughly 18–27 °C for air at the equipment inlet; the values given by the equipment manufacturer should also be taken into account. Over-cooling the room usually increases energy use without any extra benefit.
In a rack, air enters at the front and leaves at the back. Position racks so the front faces the cool-air side and the back faces the hot-air side. Close empty rack spaces with blanking panels so hot exhaust air does not circulate back to the front of the equipment. Tangled cables behind the rack and cardboard piling up in the room disrupt airflow.
In a room that relies on a single air conditioner, a failure means a direct risk of downtime. Even if a second unit is not possible, install a temperature sensor with alerts, so the problem is noticed when it happens rather than the next morning.
Water, humidity and fire
Check whether water pipes run above the room's ceiling, where the air conditioner's drain line goes, and whether the room is somewhere that could flood, such as a basement. Place floor-level water leak sensors at the risky points, and put equipment in a rack or on a raised shelf rather than directly on the floor. If there is a window, block direct sunlight.
Humidity needs watching too: very humid air can cause condensation, very dry air static electricity. Most temperature sensors also measure humidity.
A smoke detector is a basic requirement. Water-based extinguishing systems can cause further damage to electronic equipment, so the right extinguishing method for a server room should be assessed with a fire safety specialist, following the building's fire safety rules. Place a fire extinguisher rated for electrical fires next to the door and show staff where it is.
Physical access
Someone who can physically touch a server can get around many software protections: they can remove a disk, restart the machine or plug in a USB device. Keep the door locked and know who holds the key or card. With a card access system, cancelling a departing employee's card is far easier than changing a lock because of a lost key.
Track visits by cleaning staff, air conditioning technicians or external support teams with a simple logbook or the access system's records, and accompany them where possible. Use the racks' own locks, and don't leave sessions open on server consoles.
Don't use the server room as a storeroom. Paper, cardboard and cleaning supplies add to the fire load, block airflow and create new reasons for people to go in.
Cabling and labelling
Label both ends of every cable: for power cables, which UPS or socket group they go to; for data cables, which port. Run power and data cables along separate paths. Photograph the tidied rack and add the picture to your documentation, so someone being guided over the phone during a fault can find the right cable.
Monitoring and regular checks
Looking into the room every day is not realistic; instead, let the problems find you. Temperature and humidity sensors, water leak sensors and UPS event notifications (switching from mains to battery, a failed battery test, overload) should reach more than one person by email or text message. For everything else, a simple schedule is enough.
| Frequency | Check |
|---|---|
| Weekly | Load and battery status on the UPS display, warning lights, items piling up in the room |
| Monthly | Temperature and humidity logs, air conditioner filter and drain, door access records |
| Every three or six months | UPS battery test, a controlled shutdown test on one server, key and card list |
| Yearly | Battery age and replacement plan, electrical installation and earthing, fire extinguisher and air conditioner servicing |
Checklist
- Is it documented which breaker, socket and UPS feeds each device?
- Are the cables of dual-power-supply devices connected to different sources?
- Are the modem, firewall and network switches also connected to the UPS?
- Is the UPS's actual load known in watts, and has a controlled shutdown been tested?
- Are the battery installation dates and replacement times known?
- Does the air conditioner run continuously and restart by itself after a power cut?
- Do temperature, humidity and water leak alerts reach more than one person?
- Is it known who can enter the room, and are visits recorded?