Three air conditioners in a server room: arranging a duty pair and standby unit

Consider a design example for a telecoms server room: three Mitsubishi Electric air conditioners, each rated at 10 kW, with design heat gains of around 10 kW. All units have the same setpoint and are enabled to operate simultaneously.

A complex model need not be the starting point for this facility. The first step is to agree on the duty pair, standby unit and temperature monitoring at the required locations. The following describes a possible control sequence, not the results of tests already carried out.

First, check how much cooling is actually available

Three nominal ratings of 10 kW do not establish three independent, fully capable backup provisions. Sensible cooling capacity under operating conditions, air distribution, seasonal limitations and shared power and control components must be checked.

If two units are required to meet the design duty, the third can provide N+1 redundancy with an appropriate system architecture. If one unit fully and reliably covers the load, the installed capacity may correspond to a different level of redundancy. The arrangement of ‘two running, one on standby’ does not by itself determine the reliability class of the whole facility.

For server equipment, air conditions at its inlet matter, not just the average room temperature. ASHRAE specifically addresses sensor placement and airflow mixing in data centres and telecoms facilities. ¹

How rotation works

Let us label the units A, B and C. In normal operation, A+B run while C is ready to start. After an agreed interval or difference in running hours, the combination changes to B+C, then A+C.

The changeover must not begin by switching off a healthy duty unit. First, issue a permitted command to the incoming unit, check its status and confirm that temperatures are sufficiently stable. Only then place the other unit on standby. Specific delays must suit the equipment.

If the incoming unit does not confirm that it has started, rotation is cancelled and the operational pair remains in service. At a critical moment, reliability matters more than equal hour-counter readings.

Temperature monitoring must detect hot spots

For this example, assume two primary measurement points near representative rack air inlets and a third check point. Final placement is determined by a site survey, not by symmetry on a drawing.

Readings should not be blindly averaged, concealing local overheating. Equally, a single faulty sensor must not dictate the entire operation without validation. Separate logic is needed to assess data validity and respond to loss of measurements.

The standby unit starts when an agreed temperature deviation occurs, a duty unit fails or a required confirmation is missing. The emergency sequence must run locally, without waiting for a response from a cloud model.

What ‘one-third fewer hours’ means

In an idealised operating scenario where exactly two units are enabled simultaneously throughout the year, the total enabled time is:

2×8760=17 520 unit-hours

With an equal distribution, each unit accounts for 5840 hours instead of 8760 — 33,3% less time in an active role.

This is not a proven reduction in compressor running hours, a 33% increase in service life or a one-third saving in electricity. Compressors modulate or cycle off; actual running hours and consumption must be assessed separately.

Where the value lies, without extra equipment

The control sequence described does not necessarily require three electricity meters and rack-level metering from the outset. It needs reliable temperatures, status signals and a dependable interface for permitted control commands. Electricity metering is added if energy efficiency itself needs to be assessed.

Rotation and emergency sequences are tested to an agreed programme, without switching off all cooling to ‘study the rate of temperature rise’. Functional testing is part of commissioning. ²

The outcome is not algorithmic complexity, but clear standby readiness: which units are running, which will take over the load and what happens if a unit fails.