Server room cooling

Sizing from the IT load

The useful simplification: essentially all the electrical power drawn by IT equipment is converted to heat in the room. If the racks draw a given number of kilowatts, that is the heat load, continuously, regardless of the weather.

So the sizing starts with the actual electrical load – measured at the distribution board or the PDUs, not estimated from equipment labels, which are almost always badge ratings far above real draw. Add the UPS losses, the lighting, any solar and fabric gain if the room has an external wall or window, and the occasional person.

Two additional questions matter:

• What is the growth plan? A room sized exactly to today’s load has no headroom, and adding capacity later is more disruptive than providing it now.
• What is the acceptable room temperature? Modern equipment tolerates considerably warmer conditions than the traditional assumption, and running a comms room unnecessarily cold is a large and continuous waste. Check the equipment manufacturers’ stated operating ranges rather than defaulting to a number somebody remembers.

Redundancy that actually works

The question is not whether the cooling will fail, but what happens when it does.

Duty rotation deserves emphasis. A standby unit that sits idle for two years will very likely not start when it is finally needed – seals dry, condensate paths block, and nothing has proven otherwise. Rotating duty weekly means both units are continuously proven and both wear evenly.

Monitoring and alarms

A cooling failure at two in the morning is only a disaster if nobody knows about it until nine. Monitoring is the cheapest resilience available.

1. Room temperature sensors, ideally more than one – at the rack inlets rather than on a wall by the door, because that is the temperature the equipment actually experiences.
2. Alarm on high temperature, set at a level that gives time to respond rather than at the point of damage.
3. Alarm on unit failure, from the units themselves.
4. Water detection under the units and in the condensate path. A condensate leak into a rack is a different kind of disaster from overheating.
5. Remote notification that reaches a person – text, email, or through an existing monitoring platform. An alarm that only sounds in an empty building is decoration.
6. Escalation, so an unacknowledged alarm reaches somebody else.
7. Test it. An alarm system nobody has tested since installation is an assumption.

Where the room is important enough, tying the alarms into the organisation’s existing IT monitoring is usually easier than a separate system, because that is the platform people already watch.

The rest of the room

• Airflow. Cold air must reach the rack inlets and hot exhaust must return to the cooling unit. In a small room this mostly means not blocking the front of the racks and not letting hot exhaust recirculate. In a larger room it means thinking about hot and cold aisles.
• Rack orientation. Racks facing each other front-to-front means cold aisles; back-to-back means hot aisles. Racks all facing the same way means every rack breathes the previous one’s exhaust.
• Blanking panels. Empty rack U-spaces let hot air bypass straight back to the inlet. Blanking plates cost very little and measurably reduce inlet temperatures.
• Condensate. Gravity drainage, away from anything electrical. Where a pump is unavoidable, it needs its own failure alarm.
• Sealing the room. Cooling a room whose door is propped open cools the corridor.
• Humidity, which matters less than it used to for modern equipment but is worth knowing about in a room with a wide temperature swing.
• Access for maintenance, which in a room full of racks is frequently an afterthought.