Comms Rooms: A Building Services Engineer's Guide

In this article we will look at a few things we encounter when designing the cooling system of a Comms room or a Server room. Most of the time, the terms used by the IT team are ambiguous and cause us trouble while sizing the air conditioning system. Let’s look at some of the terms and basic architecture of a Comms room.

1. Room Hierarchy and Spatial Constraints

The first thing we must understand is the architectural hierarchy of IT spaces. This hierarchy dictates where we place cooling plant and how we route services.

• Main Comms Room (MDF / MER / BD) This is the core of the network. It carries the highest continuous heat loads and runs 24 hours a day, every day of the year. We therefore normally provide precision cooling with N+1 redundancy and gas fire suppression.

• Floor Comms Room (IDF / SER / FD) These rooms distribute data to local zones on each floor. Architects usually stack them vertically to create a continuous IT riser or we should ask them to do so. This helps us with our Mechanical and Electrical services routing.

There is one absolute physical limit that overrides almost every other consideration: the 90-metre rule for permanent copper links. Beyond 90 metres, copper data signals degrade to the point where the connection becomes unreliable. This physics constraint fixes the location of every IDF. We cannot simply move a comms room closer to a plant room to shorten chilled-water pipework.

Inside these rooms the hardware itself imposes further spatial rules.

• An enclosed cabinet (rack) forces strict front-to-back airflow. We must deliver cool air to the front doors and extract hot air from the rear.
• One rack unit (1U or 1RU) measures exactly 44.45 mm. A typical 42U or 47U cabinet stands roughly 2.2 m high. All high-level mechanical services, cable trays, and pipework must be coordinated in section drawings so they never clash with these cabinets.

2. Hardware Components and Thermal Profiles

Not every rack produces the same heat. We treat each rack according to its actual thermal profile rather than applying a single average figure.

• Server racks typically dissipate 3.0 to 10.0 kW. Their load is highly dynamic and peaks with compute demand.
• Network racks produce 0.5 to 3.0 kW of steady heat. They are driven largely by Power over Ethernet. We shall be careful about the rating in this case since if it is POE, only a fraction becomes a cooling load in the comms room.
• CCTV and storage racks run almost flat at 1.0 to 3.0 kW with a near 100 percent duty cycle. We apply a diversity factor of 1.0.
• AV racks generate 0.5 to 2.0 kW of intermittent heat. We size cooling on the 1/8th power rule rather than the nameplate rating.
• Controls racks stay below 0.5 kW.

Now let us look at the network hardware that lives inside these racks.

Patch panels are completely passive. They terminate the solid-core horizontal cables and generate zero heat. Short stranded patch cords then connect the panel to the switch. This arrangement protects the permanent cabling from wear and makes changes easy to implement.

Access switches (Layer 2) do the real work. They route traffic according to MAC addresses and also supply Power over Ethernet to cameras, wireless access points, and other field devices. Power is therefore drawn both by the switch’s own electronics and by the devices it feeds.

Data flow follows a clear pattern. Twenty-four horizontal copper cables typically feed into twenty-four switch ports. The switch buffers and aggregates that traffic, then sends it upstream through two to four high-speed fibre uplinks to the core (Layer 3). The system relies on a contention ratio, often around 20:1, because not every device transmits at peak rate at the same moment.

Copper is used for the final field runs under 90 metres because it can carry both data and power. Fibre is used for risers and uplinks. Fibre carries photons, so it is immune to electromagnetic interference, supports far higher bandwidth over long distances, and keeps containment sizes small.

The physics of heat generation is straightforward once you see it. Switches and servers perform no mechanical work. Their power supplies convert 230 V AC into low-voltage DC. As those DC electrons move through the silicon lattice of the microchips, they collide with silicon atoms. The collisions transfer kinetic energy and cause the atomic lattice to vibrate. Those vibrations are what we experience as sensible heat. In practical terms, virtually 100 percent of the electrical power that enters the equipment leaves as heat.

3. The Nameplate Fallacy: Real Power versus Electrical Schedules

If we size cooling plant from electrical nameplates or fuse ratings, we almost always oversize the system.

Consider a server fitted with two 1000 W power supplies. In normal operation the server may draw only 300 W in total (150 W from each supply). If one supply fails, the remaining supply simply takes the full 300 W. The heat load is therefore 300 W, not 2000 W.

The mechanical consequences of oversizing are serious. An oversized direct-expansion unit short-cycles its compressor, destroying the bearings. Even when the thermal load is low, the fans still push the design mass flow rate. The temperature difference across the coil collapses according to the familiar equation:

$Q = \dot{m} \cdot C_p \cdot \Delta T$

When $\Delta T$ becomes too small, return air temperatures can drop low enough to freeze the cooling coil. The only reliable approach is to design from manufacturer-calculated thermal dissipation data and usually never from electrical schedules.

4. Cooling Strategies and Psychrometrics

Main Comms Rooms (MDFs) need precision close-control units (CCUs or CRAHs). The reason is the sensible-heat-ratio problem. Comms rooms produce almost no moisture, so the sensible heat ratio is essentially 1.0. A standard comfort-cooling coil that evaporates at around 6 °C will strip moisture from the air, drive relative humidity down, and create a serious risk of electrostatic discharge damage to silicon.

We solve this by raising the chilled-water temperature (for example to 14 °C) so that the coil surface stays above the room dew point. Raising the water temperature, however, reduces the log-mean temperature difference:

$Q = U \cdot A \cdot \Delta T_{\text{LMTD}}$

Close-control units compensate for the lower $\Delta T_{\text{LMTD}}$ by using very large cooling coils (high surface area $A$) and high-volume EC plug fans (high mass flow $\dot{m}$). The result is 100 percent sensible cooling while the central chillers run at a much higher coefficient of performance and enjoy many more free-cooling hours.

Floor Comms Rooms (IDFs) are different. Heat loads are lower and more constant, space is tight, and the equipment is more resilient. Standard fan-coil units or DX split systems are usually acceptable. Although the coil still operates below dew point, severe over-drying is prevented by two natural mechanisms:

1. Moisture infiltrates from the surrounding occupied offices and equalises vapour pressure.
2. Two-way pressure-independent control valves modulate. As load falls the valve closes, coil temperature rises above dew point, and latent cooling stops.

We still apply strict rules: size the unit on sensible capacity only, provide gravity-drained deep secondary drip trays, and feed the unit from a true 24/7 cooling source (dedicated DX or a continuous chilled-water loop).

5. Redundancy, Resilience, and Maintainability

Cooling failure in a densely loaded room can produce unwanted temperature rise within 5 to 15 minutes. We therefore design to the N+1 principle. N is the capacity required to meet the peak load. N+1 means that capacity plus one complete standby unit.

Eliminating single points of failure requires attention to every system:

• Pipework: N+1 close-control units connected to a single header still form an N-level system. We use dual-path (A/B) headers or isolation ring mains.
• Electrical: Redundant cooling units must be fed from separate distribution boards and separate generator paths.
• Controls: Units operate on local, hardwired master/slave networks. The central building management system monitors only; it never controls the cooling sequence.

6. Fire Detection and Suppression Interfaces

Water sprinklers destroy IT equipment, so comms rooms use gas suppression. The mechanical engineer largely decides whether that suppression system succeeds or fails.

1. Detection must use aspirating smoke detection (ASD or VESDA). These systems sense the earliest off-gassing from overheating cables. Ordinary point detectors are almost useless in the high-airflow environment of a modern IT room.

2. On second-stage fire alarm a hardwired interlock must instantly cut power to every cooling unit in the room and close all motorised fire and smoke dampers. If plug fans continue to run they will dilute and disperse the suppression gas, preventing extinguishment.

3. Gas discharge creates a rapid pressure rise. Counter-weighted pressure-relief dampers sized for the discharge volume must open to protect the room structure, then snap shut again to retain the gas.

4. The room must maintain the required gas concentration for a full ten minutes. Every mechanical penetration (duct, pipe, or cable tray) must be hermetically fire-stopped.

5. After discharge a dedicated mechanical extract system, started manually from outside the room, purges the toxic by-products and residual gas directly to atmosphere.

This is just the basics of Comms room and a good dialogue with the IT team is essential to size the cooling equipment correctly.