How long can data center cooling ride through an outage? A thermal-buffer calculation

How long will cooling hold after power fails?

Define ride-through by the first unacceptable equipment temperature, flow, pressure or performance. Time to throttle, alarm, shutdown and damage are different endpoints; name the one the project must prevent.

A chiller outage with distribution running leaves a path from the racks to stored cold water. A stopped cold-plate pump removes that path at the component, even when the remote tank is full. Schneider Electric's WP179 Rev 0 shows the corresponding limitation for chilled-water CRAHs: stored water becomes useful cooling when water circulation and airflow are available. Its room models distinguish bulk temperatures from local rack conditions.

OCP Liquid Distribution Guidance, revision 1.0, January 2024 defines the liquid-loop boundaries.

Use liquid-cooling redundancy to establish surviving paths and generator acceptance to establish electrical recovery. Their shared acceptance record needs the thermal timeline calculated here.

For ice or phase-change storage, use the supplier's discharge capacity and power curve at the required outlet temperature. A sensible-water calculation alone does not describe latent storage.

How do you calculate the thermal buffer?

Multiply accessible coolant mass by its specific heat and permitted warming to find usable stored energy. Divide that energy by the net heat accumulating in the coolant for a constant deficit; integrate the deficit when plant capacity changes. OCP Modular TCS v1.0, August 25, 2025, Appendix H uses this thermal-mass method.

Choose the water boundary before assigning heat load. Include heat entering that node from the equipment and relevant pumps; allocate residual air-cooled rack heat separately. Outdoor condenser rejection and compressor input are not automatically part of chilled-water duty. Count auxiliary heat separately only when it is absent from the stated load.

The following method uses OCP Modular TCS v1.0, August 25, 2025 water properties and DOE EnergyPlus Engineering Reference 24.1, Water Thermal Tanks mixed-node energy accounting. Energy is in kJ, power in kW, time in seconds, volume in m³, density in kg/m³ and specific heat in kJ/(kg·K).

For logged data, average consecutive deficit readings, multiply by their interval in seconds, and accumulate the resulting kJ. Track the maximum depletion, which can occur before full plant recovery. At full recharge, reset depletion to zero and discard further surplus; begin the next event from that stored-energy state.

Constant loss versus a recovering plant

Credit cooling where it reaches the modeled boundary, using synchronized flow and supply/return temperatures. A compressor command earns no thermal-power credit. An economizer contributes only while its complete flow path and ambient-dependent duty survive; include transport delay once, in the delivered-capacity curve.

The operating supply/return delta-T transports heat; the permitted bulk-water warming consumes storage. Keep those temperature differences separate, and distinguish kW of duty from kWh of stored heat.

The calculations use OCP’s rounded water inputs: density 1,000 kg/m³ and specific heat 4.186 kJ/(kg·K). NIST Chemistry WebBook SRD 69, water, Chase 1998 documents temperature-dependent heat capacity, but its displayed liquid fit starts at 298 K; the 14°C example uses OCP’s approximation. A mixed water node predicts bulk water temperature; branch mixing, coil response and room recirculation require coupled transient models to predict equipment inlets.

Which water volume actually counts?

Count water accessible through the surviving flow path before equipment reaches its limit. Reconcile filled volumes, valve lineup and initial temperatures to the discharge path. The ledger contains 16 m³ physically but credits 10 m³ effectively.

Anvilfield's selected tank availability factor is 75%; it represents participating volume in this lumped calculation. Use the same credited mass consistently in the energy and temperature calculations. Establish the installation's factor from its temperature profile and outlet discharge performance. The credited tank water and cold supply piping both start at 14°C, with a 19°C cutoff.

Keep facility-water system (FWS) inventory separate from technology-cooling system (TCS) inventory across the CDU heat exchanger. The ledger credits facility-side storage only. Count each physical volume once; a reserve branch behind closed valves contributes nothing during the modeled interval.

Warm return water receives no blanket 14°C credit. Calculate each connected mass's remaining capacity to its applicable cutoff, and resolve the mixing when warmer inventory rejoins. Credit equipment metal only where a transient model or test establishes that it exchanges useful heat during the event.

Mixed buffers and stratified tanks need different temperature models

DOE EnergyPlus Engineering Reference 24.1, Water Thermal Tanks separates a uniformly mixed tank from a stratified tank represented by coupled temperature nodes. A mixed tank warms throughout; a stratified tank can maintain a cold outlet until the thermocline reaches it. The same total water inventory can therefore produce different outlet-temperature histories.

Determine useful energy at the outlet cutoff from a discharge test or stratified model. An already measured usable-energy value needs no second efficiency discount. The chilled-water thermal storage guide covers tank diffusers, construction and charging arrangements.

What inlet temperature envelope must the buffer protect?

Use the rack or server model’s operating and performance limits at the specified measuring point. Rack air inlet, FWS inlet, TCS inlet and chip temperature are separate boundaries. ASHRAE, Emergence and Expansion of Liquid Cooling in Mainstream Data Centers, 2021 explains the thermal resistance between coolant and component; a water classification does not establish chip temperature.

The table uses ASHRAE Thermal Guidelines, fifth edition revised and expanded, reference card ©2021/2024. Its allowable air ranges permit thermal management within the band, so acceptance also needs the project’s required IT performance. Humidity, noncondensation and altitude remain part of the equipment envelope.

Calculate each air-inlet excursion as the maximum minus minimum within the moving window, including the fall during recovery. These air criteria do not establish a liquid-water ramp limit; tape equipment has separate, narrower provisions in the cited ASHRAE card.

The ASHRAE TC 9.9 committee index, checked November 5, 2026 says the Datacom Encyclopedia replaced the book series and introduces separate S Classes for TCS supply. Use the equipment’s declared class and referenced edition; W32 in the cited card concerns facility water. The thermal-envelope guide provides broader setpoint context.

Translate water warming into a rack limit

The selected 3°C CDU approach temperature translates the 22°C TCS ceiling into a 19°C FWS ceiling. From the 14°C FWS start, usable warming is 5°C. Verify approach under the event’s flow and load.

If approach grows to 5°C, the same TCS ceiling lowers the FWS ceiling to 17°C and leaves only 3°C of warming. Exchanger fouling or reduced primary flow can consume storage margin without changing tank volume. OCP Liquid Distribution Guidance revision 1.0 (2024) addresses condensation during transitions: keep exposed surfaces above local dew point or apply the engineered condensation controls, including during recharge.

How do generator, pump and valve restart times fit together?

Timestamp bus power, controls, valves, proven flow, first useful cooling and restored duty from the same power loss. Overlap parallel tasks; add only serial dependencies.

Distribution and TCS circulation remain available throughout this timeline; restarted pumps serve the chiller source side. Lost distribution requires a separate local survival assessment.

Prove flow, not just a run command

A running motor with a closed valve, air-bound pump or blocked strainer has not restored cooling. Capture branch flow, differential pressure, VFD state, valve feedback and permissives. Chilled-water pumping arrangements determine the surviving duty point; CDU commissioning establishes performance on both exchanger sides.

Using the OCP water approximation, Anvilfield estimates that transporting 1 MW at a 10°C operating loop differential requires 23.89 kg/s, approximately 23.89 L/s; at 100 kW it requires 2.39 L/s. Actual duty must also satisfy OEM minimum flow, available pump head and pressure-drop limits.

Build the hydraulic dependency path from detection, power and control availability, valve travel, pump acceleration and flow-proof dwell. Verify the backed power path for sensors, controls and required air movers. For a branch initially holding warmer water, divide that segment's volume by actual volumetric flow to estimate displacement time, then resolve mixing and exchanger response.

Convert the energy budget into a latest useful-cooling start

Anvilfield estimates a latest zero-reserve cooling-ramp start of 149.30 seconds after failure for 209.30 MJ of storage, a 1 MW load and a linear 120-second recovery ramp to 1 MW. The ramp consumes 60 MJ, leaving 149.30 MJ for the preceding zero-cooling interval. Starting at the planned 60 seconds leaves 89.30 seconds of additional delay capacity.

Deduct the required energy reserve and uncertainty allowance before setting restart acceptance. If rack flow is lost, its validated local survival interval also governs. Faster electrical transfer improves the thermal result when it advances useful cooling along the actual dependency path.

What do manufacturer restart claims actually mean?

Manufacturer restart claims describe specific products, options and starting conditions. Trane RTAC-SLB001-EN, printed 07/04/2020 and Vertiv AFC SL-71334GS_REVA_08-24, section 2.10 use clocks beginning at restored power; add the preceding facility interruption when comparing them with an outage budget.

Vertiv’s August 2024 revision A fast-restart procedure applies to order-code digit 17 options 1, 2 or 3 with controls kept operational. The controls buffer powers controls, not the pumps or fans; the specification supplies separate restart steps rather than a guaranteed time to full thermal duty.

Procure the supplied model, options, refrigerant, controller revision and recovery curve together. Record inhibits and permissives; witness delivered capacity under the specified ambient and condenser conditions.

Worked example: can 10 m³ bridge a 1 MW outage?

The worked recovery uses 120 MJ of the available 209.30 MJ, raises modeled water temperature by 2.87°C and leaves 89.30 MJ at 180 seconds, Anvilfield estimates. Its physical path is buffer and cold supply piping, maintained facility distribution, CDU exchanger, separately maintained TCS circulation, then racks. Source pumps and chillers recover on the plant side of the buffer.

The load is 1 MW entering the facility-water boundary. Electrical recovery takes 30 seconds; useful cooling starts at 60 seconds and reaches full duty at 180 seconds. Storage credits water only.

The modeled TCS peak stays below its selected ceiling. Check residual air-cooled rack load against its separate inlet envelope.

Check discharge power as well as stored energy

The store and connecting hardware must deliver the initial 1 MW shortfall. Anvilfield’s water heat-transport calculation requires 23.89 L/s at a 10°C operating differential, or 47.78 L/s at 5°C. Use actual tank-port and branch temperatures for those duties; the allowed bulk warming is a separate input.

Verify port and diffuser capacity, valve capacity, pump head at the surviving duty point, bypass mixing, exchanger duty and OEM minimum flow. An energy inventory becomes useful cooling only through a path capable of carrying the required power.

Convert the result into required storage

Convert the effective-volume requirement in Table E to physical tank volume by subtracting credited piping and dividing by tank availability. Apply the allowance before this conversion and calculate before rounding. The selected 12 m³ tank supplies additional reserve; the 20% allowance is a design choice.

Method: Anvilfield estimates use the water-property approximation in OCP Modular TCS v1.0, August 2025, the stated accessible inventory and temperature limits, and trapezoidal integration of the cooling-recovery curve. Heat load includes heat entering this water boundary; other thermal masses receive no credit. Tank availability, timing, approach and the energy allowance are stated inputs, with sensitivity calculated by changing one input at a time.

Which changes consume the ride-through margin fastest?

Warmer starting water, inaccessible inventory and wider exchanger approach consume margin. Solve for the first time accumulated deficit reaches usable energy, including depletion after restart.

The reduced-volume and warmer-start cases fail during the recovery ramp. The higher-load case continues consuming storage after nominal recovery because delivered duty remains below its new heat load. Test credible combinations as well: warm starting water, reduced inventory and a slower restart can occur together.

Anvilfield estimates that each additional 60 seconds at a 1 MW deficit consumes 60 MJ, requiring another 2.87 m³ of effective water at 5°C warming. Use this relationship to compare added storage against a verified reduction in recovery time. Storage readiness must report temperature and available energy, not just a full tank.

Glycol and uncertainty change the inputs

OCP Modular TCS v1.0 Appendix H distinguishes water from glycol mixtures. Use the actual coolant product, concentration and temperature for density and heat capacity, and its viscosity for hydraulic calculations. A universal glycol percentage penalty obscures both energy capacity and flow performance.

The table’s selected 0.5°C uncertainty allowance removes 20.93 MJ of reserve; it is an input sensitivity, not a sensor-accuracy claim. Use the recovery curve at adverse design conditions, including hot-weather condenser performance, cold-weather viscosity and economizer transitions. Low operating delta-T affects transport duty and must remain distinct from the storage warming margin.

What changes for direct-to-chip liquid racks?

Local circulating volume must be matched to local heat load. The opening's small-loop result assumes all credited coolant remains accessible; a stopped cold-plate loop needs an equipment transient rather than the central tank duration.

OCP Modular TCS v1.0 Appendix H’s different inputs, 100 liters, 10 kW and 15°C warming, yield 627.90 seconds, or 10.47 minutes. Both calculations require accessible circulating volume.

For a stopped cold-plate loop, use server transient data or a validated component model that includes surviving flow and thermally coupled materials. Establish time to throttle and shutdown at the actual starting condition. Direct-to-chip architecture identifies the FWS, CDU and TCS boundaries that must remain separate.

Surviving TCS circulation retains its own thermal inertia during a facility-side upset. Preserve airflow for components outside the liquid heat-capture fraction. Power capping earns thermal credit only when allowed service level, command latency, achieved reduction and control-power continuity are recorded.

The load-reduction row in Table G peaks before full plant recovery because cooling overtakes the reduced load. Record the achieved heat reduction at the modeled boundary; an electrical cap does not instantly establish the same reduction in water-side duty.

How soon is the buffer ready for a second outage?

Replacing the worked example's 120 MJ of consumed storage takes 8 minutes at 250 kW of net spare cooling capacity, Anvilfield estimates, so a successful first ride-through does not establish readiness for an immediate repeat event. That recharge duty requires 1.25 MW of delivered cooling while the original 1 MW load continues.

Spare capacity becomes available at 180 seconds in the selected recharge scenario. Full recharge takes another 480 seconds, reaching the initial state at 660 seconds, or 11 minutes from the first failure. Cooling equal to load merely stops depletion.

Readiness for one repeat event arrives earlier than full recharge. At 180 seconds, 89.30 MJ remains; rebuilding reserve to the next event's 120 MJ requirement needs 30.70 MJ. At 250 kW spare duty, Anvilfield estimates readiness at 302.80 seconds from the first failure, or 398.80 seconds for the selected 144 MJ requirement including the 20% allowance.

These repeat-event times assume the same load, usable inventory, approach and delivered-cooling recovery curve as the first event. A warmer restart, new inhibit or different valve state changes the required energy. For the same mixed 10 m³ node, Anvilfield estimates a readiness temperature no higher than 16.13°C for the base event, or 15.56°C with the allowance.

Tie the energy-ready indication to actual temperature or stratified profile and the next-event duty. Test retransfer and repeat interruptions at the owner's required separation; prove the required discharge power as well as stored energy.

How do you prove cooling continuity in a field test?

Witness the integrated test sequence at the specified thermal load and hydraulically adverse operating configuration. Synchronize electrical events with measured thermal duty. A generator load bank rejecting heat outdoors does not impose the rack or CDU cooling load; the test heat must enter the intended cooling path.

Measure source and load-side temperatures and flow, buffer profile, rack air inlets, TCS inlets and IT performance. EPA ENERGY STAR, Use Sensors and Controls, undated supports sensing equipment conditions rather than relying on HVAC return-air control. Stable central water can coexist with a hot rack when containment or local airflow fails.

Anvilfield proposes 1-second central-loop logging and 100-millisecond acquisition for an expected seconds-scale pump handoff. These are test settings; document sensor response and calibration, and select faster instrumentation for faster events. A fast logger cannot recover a transient that a slow sensor averages away.

Field checklist

• Identify the failure boundary, affected racks and first unacceptable condition.
• Record rack models, declared classes, temperature, flow, pressure and performance limits.
• Allocate liquid and residual air heat loads; identify auxiliary heat inside each boundary.
• Reconcile filled volumes, starting profiles, valve lineup and accessible storage to drawings.
• Record coolant product, concentration and calculation properties; separate FWS and TCS nodes.
• Confirm available redundancy and power continuity for controls, pumps, sensors and air movers.
• Record chiller options, controller revision, permissives and restart inhibits.
• Prove branch flow, pump head, exchanger duty, tank discharge power and actual valve position.
• Synchronize logs and document sensor calibration, response and acquisition interval.
• Set project abort thresholds ahead of equipment limits and name the operator responsible for restoring cooling or reducing load; apply the approved heat distribution and authorized fault sequence.
• Capture inlet extrema, flow minimums, pressure and IT status through recovery.
• Integrate measured deficit; compare peak depletion and the first threshold crossing with the model.
• Demonstrate retransfer and the specified repeat event from recorded initial conditions.
• Prove recharge and the energy-ready indication; retain settings, defects and retest evidence.

Acceptance uses the complete event and the worst measured branch

Each specified temperature, flow, pressure, rolling excursion and performance criterion must hold through loss, restart, recovery and retransfer. Retest after corrections and after changes affecting load, coolant, controls, storage or hydraulic sequence. Set periodic recurrence in the owner/OEM maintenance plan.

The engineer sets design criteria, OEMs set equipment limits, the owner sets service objectives, and the commissioning lead coordinates evidence. Uptime Institute's July 20, 2017 explanation of Tier IV continuous cooling calls for a stable thermal environment through full mechanical restart and rated-load recovery. Use the project's applicable Tier criteria and recovery evidence to define duration.

What should a contractor include in the test scope?

Define the witnessed event, thermal boundary and report deliverables before assigning labor. Anvilfield estimates 48 labor-hours for one planned central-loop event at an existing commissioned plant using the scope below. The owner supplies the test heat load and authorized operating staff; quantities are selected planning inputs dated November 5, 2026.

Method: add the selected labor quantities. Assign instrument rental, heat-load equipment, switching crews, OEM attendance, temporary power or cooling, water treatment, access, travel, repairs and repeat tests separately. Staff-hours are not elapsed outage duration; scheduling depends on access windows and parallel crew assignments.

Carry controls exports and reporting alongside the witnessed window. Separate tank equipment, piping, supports and controls installation from event verification. Estimators tracking cooling-plant work find agency solicitations and early budget and capital-plan signals through PorchRocket Bid Control's public HVAC bids.

What belongs in the cooling-continuity turnover record?

Deliver the filled volume ledger, P&ID, valve lineup, fluid properties, heat-load allocation and model-specific limits with manual revisions. Retain restart settings, source and load timestamps, recovery and recharge traces, sensor calibration and uncertainty. Record the starting storage condition and test configuration beside each result.

The acceptance matrix should pair each criterion with its measuring point, worst recorded value and disposition. Explain model-versus-test differences, retain final controls files, and identify unresolved defects and successful retests. Retain the readiness threshold used by controls.

Retain raw time-series exports for recalculation after load or recovery changes. Make the controlling restriction visible to operations: maximum tested load, minimum reserve, required operating paths and permitted repeat-event separation.

Sources checked and update

2026-11-05: Reviewed source editions, inlet limits, manufacturer restart sequences, energy calculations and internal links. Clarified repeat-event readiness and controls power; replaced budget prices with a labor-scope worksheet. The source list identifies the EnergyPlus and Schneider document mirrors.

How to cite this page: Anvilfield, How long can data center cooling ride through an outage? A thermal-buffer calculation, 2026, https://anvilfield.com/field-guides/datacenter/data-center-cooling-ride-through-thermal-buffer/

Related tools

FAQ

People also ask

• What is an NFPA 110 acceptance test?
• What is thermal energy storage in a data center?
• What is concurrent maintainability in liquid cooling?
• What is a CDU in a data center?
• What is the ASHRAE recommended temperature for a data center?

Sources checked

• OCP Modular TCS v1.0, August 25, 2025 Appendix H: thermal mass, water properties and local-loop example. Checked 2026-11-05.
• OCP Liquid Distribution Guidance, revision 1.0, January 8, 2024 FWS/TCS boundaries, flow continuity and distribution. Checked 2026-11-05.
• ASHRAE Thermal Guidelines, fifth edition revised and expanded, reference card ©2021/2024 Air-inlet classes, altitude, rolling excursions and fixed-edition facility-water classes. Checked 2026-11-05.
• ASHRAE, Emergence and Expansion of Liquid Cooling in Mainstream Data Centers, 2021 Coolant-to-chip thermal resistance and liquid-cooling boundaries. Checked 2026-11-05.
• Trane RTAC-SLB001-EN, printed 07/04/2020 Specific RTAC complete-power-loss restart sequence; date retained as printed. Checked 2026-11-05.
• Vertiv AFC SL-71334GS_REVA_08-24, section 2.10 Section 2.10, page 6: optional controls power and fast-restart sequence. Checked 2026-11-05.
• DOE EnergyPlus Engineering Reference 24.1, Water Thermal Tanks DOE-authored reference hosted by Big Ladder Software; mixed and stratified tank energy models. Checked 2026-11-05.
• NIST Chemistry WebBook SRD 69, water, Chase 1998 Temperature-dependent liquid heat capacity; displayed fit range 298 to 500 K. Checked 2026-11-05.
• Uptime Institute, Tier Certification myths and misconceptions, July 20, 2017 Tier IV continuous cooling through mechanical restart and rated-load recovery; historical explanation, not a universal tank-runtime value. Checked 2026-11-05.
• EPA ENERGY STAR, Use Sensors and Controls, undated Equipment-level temperature sensing and control at actual servers. Checked 2026-11-05.
• Schneider Electric, Data Center Temperature Rise During a Cooling System Outage, WP179 Rev 0 Manufacturer-authored public PDF hosted by ECL-IPS; circulation and room-transient modeling. Checked 2026-11-05.
• ASHRAE TC 9.9 committee index, checked November 5, 2026 Current publication context and separate TCS S Classes; no numerical S-class thresholds inferred. Checked 2026-11-05.