Continuous cooling and thermal ride-through

When utility power fails in a data centre, the UPS carries the IT load instantly and the generators start. The servers never notice. The cooling system, however, usually does. Chillers, compressors and many fans are not on UPS power, so they stop until the generators come online, and then many of them take minutes to restart and return to full capacity.

During that gap, the IT equipment keeps producing heat. Whether the room stays within safe temperatures depends on how much heat it is producing, how much thermal mass is available and what parts of the cooling system keep running. This is called thermal ride-through, and for high-density facilities it can be the limiting factor in the whole design.

The timeline of a power failure

A typical sequence for a facility with standby generators looks like this:

Generator start times are well defined. Chiller restart time is the bigger unknown. A conventional centrifugal chiller may need several minutes to complete its restart sequence and reach full load after a power interruption. Many manufacturers offer rapid restart options that reduce this considerably. The actual time must be confirmed with the equipment supplier and verified in testing.

A return of utility power brings a second transfer, so the sequence can repeat. Designs should consider back-to-back interruptions as well.

How fast does the room heat up?

Without cooling, the temperature rise in an IT room depends on load density and the thermal mass of the air, the equipment and the building structure. Air alone has very little heat capacity. In a modern high-density room, the air temperature at server inlets can rise several degrees within a minute if airflow stops and hot exhaust recirculates.

Two factors make this worse:

• Airflow loss. If air handler fans stop, hot aisle air recirculates to server inlets almost immediately, and contained aisles can overheat faster than open rooms.
• IT fans speeding up. As inlet temperature rises, server fans accelerate, which can increase recirculation and power.

Ride-through is therefore not just about stored cooling. Keeping air moving is often the first priority.

Design strategies

1. Put fans and pumps on UPS power

The most effective step for many facilities is to put CRAH fans, chilled water pumps and the controls that run them on UPS power. With chilled water still circulating through coils and fans still moving air, the room uses the cold water already in the piping and coils while the chillers restart. Modern EC fans and variable speed drives handle the transfer well, but their ride-through behaviour should be confirmed.

Controls are critical. If the building automation system or chiller plant controller reboots on a power blip, it may take longer to recover than the chillers themselves. Controllers, network switches and sensors serving the cooling plant should be on UPS.

2. Use the thermal mass of the chilled water system

A chilled water loop holds a significant amount of cooling in its piping. The usable amount is:

Stored cooling (kJ) = mass of water (kg) x 4.19 kJ/kg K x usable temperature rise (K)

For water that number is easy to work out. A loop containing 20 m3 (about 5,300 US gal) of water that can warm by 5 K before supply temperature becomes unacceptable stores about 419,000 kJ, enough to absorb 1,000 kW for about 7 minutes in theory. In practice the usable fraction is lower because water mixes and coils lose capacity as supply temperature rises. Glycol solutions have lower heat capacity, which should be accounted for.

3. Add chilled water thermal storage

Where loop volume is not enough, a dedicated storage tank adds ride-through time. Stratified chilled water tanks, sized for the required minutes at full load, are common in larger facilities. A tank can be:

• In series with the supply, so it always contains chilled water ready to discharge.
• Charged and isolated, then released by control valves on a power event.
• Combined with a daily load-shifting thermal energy storage strategy, if the tank is large enough to serve both purposes without compromising the emergency reserve.

Pumps serving the tank must be on UPS. The control sequence must be simple and fail-safe, since it operates exactly when the plant is in its most disturbed state.

4. Chiller rapid restart

Specifying rapid restart features, with drives, controls and oil systems designed to return to load quickly, reduces the storage needed. Chiller sequencing must also be coordinated so that restarting machines do not overload the generators.

5. DX and in-row systems

Facilities using DX CRAC units or refrigerant-based in-row cooling have little stored cooling to fall back on. Options include putting compressors on UPS (often impractical because of size), adding short-term stored cooling, or accepting a defined ride-through time based on room thermal mass and allowable temperature limits.

6. Liquid-cooled IT

Direct-to-chip and immersion systems often need coolant distribution unit pumps on UPS, because the coolant volume near the chips is small and heat can build quickly. The facility water serving the CDUs may still rely on loop mass or storage during chiller recovery.

Setting the acceptance criteria

A ride-through design needs a clear target. Common criteria include:

• Maximum IT inlet temperature during the event, often the ASHRAE allowable limit for the equipment class.
• Maximum rate of temperature change, since ASHRAE guidance also limits how quickly temperature should change.
• The duration and number of interruptions that must be survived.

Tier or redundancy objectives also matter. Continuous cooling is expected for higher-tier facilities, as described in data center cooling redundancy and Uptime tiers.

Analysis tools

Simple energy balance calculations give a first estimate of storage volume. For more confidence, transient models track plant recovery, loop temperatures and room temperatures second by second. CFD modelling can show where hot spots develop when fans slow or stop, especially in rooms with containment.

Testing ride-through

No ride-through design should be accepted on paper alone. During commissioning, a pull-the-plug test with the data hall loaded by load banks is the most convincing evidence:

1. Load the room to the design IT load with load banks, distributed realistically.
2. Bring the plant to steady state and record baseline temperatures.
3. Open the utility breaker and let the generators, transfer switches and cooling plant respond without intervention.
4. Record server inlet temperatures, chilled water supply temperature, storage tank temperatures and chiller restart times at short intervals.
5. Restore utility power and record the retransfer.
6. Compare results to acceptance criteria and adjust sequences as needed.

The guide on load bank testing for cooling validation covers set-up in more detail.

Frequently asked questions

Is ride-through only a concern for large data centres?

No. Small server rooms with high-density racks can heat up quickly too, and they rarely have stored cooling. The same questions apply: what keeps running, and for how long can the room survive without compressors?

Does a larger UPS solve the problem?

Only if it powers the cooling equipment. Many failures occur because the UPS protected the servers but not the fans, pumps or controls that keep them cool.

Key takeaways

• Cooling usually stops when utility power fails, and chillers often take several minutes to return to full capacity.
• Keeping fans, pumps and controls on UPS power is often the single most effective ride-through measure.
• Chilled water loop volume and thermal storage tanks provide stored cooling that can be calculated with a simple energy balance.
• Set clear acceptance criteria for maximum temperature and rate of change.
• Verify ride-through with a loaded, integrated power failure test before handover.