Data hall cooling inside the envelope - and inside the MOP

Data centres are an unusual market for a BMS integrator. The energy at stake is large, the levers are well known, and the client is often more interested in how you will make a change than in what the change will save. A colocation or enterprise data hall runs around the clock, its customers have service level agreements, and an outage costs far more than any energy saving.

This article covers the BMS interface side of data hall cooling - the sensors, setpoints, fan control and sequences an integrator touches - and the procedural discipline that goes with working in a live data hall. It is not about designing cooling plant, which is specialist work. It is about running the plant that is there as well as it can be run, safely.

The efficiency gap is still there

Uptime Institute's 2025 survey put the global weighted average power usage effectiveness (PUE) at about 1.54, with little change for six years (Uptime 2025). Newer and very large sites do better; many enterprise and colocation halls do worse. Cooling is the main reason PUE sits above 1, so there is still a gap in a great many existing sites - and much of it is controls.

The large North American recommissioning dataset found a median saving of 11% in data centres - but from only four projects, so treat it as a hint rather than a benchmark.

Recommissioning savings by building type - Median whole-building energy savings, %
Lodging (hotels)3%
All building types combined (n = 446)6.4%
Data centres (n = 4)11%
Laboratories (n = 28)14%
Public order and safety16%

The lowest and highest segments and three in between - small samples (n = 4) say little on their own.

Data centres sit near the top of the recommissioning medians, but on a sample of four. North American. Source: Crowe et al. 2020 (LBNL), existing-building commissioning medians by market segment. Flags: North American, Measured.

The envelope is the headroom

The ASHRAE thermal guidelines for data processing environments (5th edition, 2021) give a recommended range of 18-27°C for air at the server inlet (Upsite). Many halls are still run far colder than that, with return air setpoints chosen years ago for older equipment. The recommended envelope is what makes optimisation safe: raising temperatures within it, at the inlet, is a well-established practice rather than a risk.

The key word is inlet. A common fault is CRAH or CRAC units controlling on return air temperature, which tells you little about what the servers are breathing. Moving control - or at least monitoring - to rack-inlet temperatures is usually the first step.

The levers, in order

A repeatable sequence, each step measurable through PUE and fan and chiller energy:

1. Rack-inlet sensing. Survey inlet temperatures at the front of racks, top, middle and bottom, across the hall. Find the hot spots before changing anything. 2. Containment and bypass air. Check blanking panels, floor grommets and aisle containment. Cold air bypassing the servers forces lower setpoints and more airflow. An academic comparison found cold aisle containment saved about 22% against open aisles (NSF-hosted paper), though containment alone does not guarantee optimum cooling. 3. CRAH fan speed control. Control fan speed on underfloor or aisle pressure, or on inlet temperature, rather than running fans at fixed speed. The fan affinity laws do the rest: at 80% speed a fan draws roughly half its full-speed power. 4. Stepwise setpoint raises. Raise supply air and chilled water temperatures in small steps, holding inlet temperatures within the recommended envelope and watching for hot spots between steps. One vendor study found raising chilled water from 7°C to 17°C cut cooling energy by 39% (DCD) - a vendor figure, and results depend on the coils and plant. 5. Economiser logic. Review free-cooling enable and changeover. In a temperate UK climate the free-cooling hours available are substantial, and changeover that hunts or locks out too early wastes them.

The EU Code of Conduct Best Practice Guidelines, published free each year by the European Commission's Joint Research Centre (JRC), give a credible, numbered list of practices to map this work against.

Faults that come up again and again

From field experience rather than a single measured dataset, the common BMS-side faults in data halls are:

Each is a sensor calibration, a sequence change or an alarm review - ordinary BMS work, done in an unusual environment.

Inside the MOP

This is where data-centre work differs most from a normal service visit. Uptime's 2025 outage analysis found power the most common cause of significant outages and cooling the second, and that 58% of outages involving human error came from staff not following procedures (Data Center Knowledge). Clients respond by controlling change tightly, and an integrator's procedural discipline sells as strongly as its controls skill.

Data-centre practice distinguishes three kinds of document:

For BMS work, every setpoint raise, firmware update, controller replacement or sequence change should have its own MOP. A good one states what will be changed and why, what must be true before starting, who must be present, the exact steps with a check after each, what success looks like in the trends, and how to put everything back if it does not. Changes go into agreed maintenance windows, and the client's operations team stays in control of the hall.

A stepwise setpoint raise fits this naturally: each step is a small MOP, with a hold period and clear criteria for the next.

The drivers behind the client's interest

Shaping the offer

A data-centre BMS service built on this might include inlet temperature surveys, sensor calibration rounds, alarm rationalisation, a library of MOP templates for routine BMS changes, a staged setpoint programme, and a quarterly trend review showing PUE and cooling energy. Keep the boundary clear: cooling plant design, electrical systems and IT white space belong to others.

Alarm rationalisation covers taming the alarm list, Fault detection on a BMS the rules a monitoring service can run, and Proving savings how to show the result. What the studies really say puts any savings figure in context.

Related tools and training: the psychrometrics, inverter (VSD) savings and alarm load tools are in tecTools, and the CHW system commissioning course is in tecTraining.

Sources mentioned: Uptime Institute Global Data Center Survey 2025 and outage analysis 2025; ASHRAE thermal guidelines (5th edition, 2021, via a published summary); academic containment study; vendor chilled water study (vendor figure); JRC EU Code of Conduct Best Practice Guidelines (2025); data centre CNI designation (2024); Climate Change Agreements (techUK summary). Results depend on the hall, its plant and its load.

More articles from 3EC

3EC logo