
Size the buffer by subtraction: take the minimum system volume the chiller data sheet requires, subtract the volume the loop already holds, and confirm the result against a run-time check at minimum load. Place the vessel on the common pump suction so the chiller and the IT load are hydraulically decoupled, and size the connecting pipe for the transient, not the steady maximum flow.
A chilled water buffer tank in a data center is a chilled-water, HVAC-duty vessel. It is not a fire-water storage tank, and the NFPA 22 fire-protection chapter map that governs fire-water duty does not apply to it; the governing references for this scope are the chiller and system design basis, ASHRAE guidance, AHRI chiller ratings and the piping code for the building services loop. The buffer's job is to give the chilled water plant a volume of water it can draw on so the chillers do not short-cycle when the IT load is low. Get the volume wrong and the most efficient part of the plant spends its life starting and stopping.
Two conventions run through the text. Volumes are quoted in cubic metres with gallons in parentheses where both appear, and the word volume always means a defined basis: nominal, usable or effective. A buffer sized on an undefined gross number is the most common defect in this scope, and it is the one this method is written to prevent.
Three different questions produce three different numbers, and confusing them is the root of most over- and under-sized buffers. The first number is the chiller minimum system volume: the total water the chiller maker requires in the circuit so the evaporator flow stays inside its limits during part-load operation. The second is the volume already in the loop from piping, coils and equipment. The third, and the only one you build a tank for, is the difference between them.
A buffer sized on the plant total answers the wrong question. The total chilled water inventory can be large without solving the short-cycling problem, because what matters is how much extra water sits between the chiller and the load at the moment the load collapses. Equally, a tank sized on a rule of thumb in litres per kilowatt ignores the chiller's actual minimum, which is read from the data sheet and varies by machine. The discipline is to name the basis of every number before you use it.
The subtraction method starts from the chiller minimum system volume, V_min_sys, read from the chiller data sheet or the AHRI-rated application guidance for the machine in question. It then subtracts the volume already present in the installed loop, V_loop, which is the sum of the piping volume from the hydraulic calculation, the coil and heat-exchanger inventory, and the equipment holding volume. The buffer volume is the remainder:
1. V_min_sys: minimum system volume the chiller requires, from the chiller data sheet (units m3 or US gal).
2. V_loop: as-built or as-designed loop volume from the hydraulic model and the coil schedules.
3. V_buffer = V_min_sys - V_loop, used only if the result is positive; a negative result means the loop already satisfies the chiller and no buffer is required.
The method is defensible because every input is traceable. V_min_sys comes off a document the chiller supplier already provides, V_loop comes off the project drawings, and the arithmetic is the audit trail. The failure mode is using a generic minimum such as a fixed litres-per-tonne figure without confirming it is the number the actual chiller states.
The subtraction method gives a volume that satisfies the chiller's minimum, but it does not by itself prove the plant will not short-cycle. The run-time check closes that gap by asking how long the buffer holds the chilled water temperature within band while the cooling load is at its minimum, Q_min. The relation is the sensible heat balance:
4. Q = rho * c * V_eff * dT / t, where Q is the load (W), rho and c are the density and specific heat of water, V_eff is the effective buffer volume, dT is the allowable temperature drift and t is the time.
5. Solve for t = rho * c * V_eff * dT / Q_min to find the hold time at minimum load.
6. Compare t against the chiller's minimum allowable run time and its minimum off-time; if t is shorter than the chiller needs, increase V_eff.
Note the use of V_eff, the effective volume, not the nominal tank volume. Dead legs, nozzles and any thermally inactive region reduce what the chiller can use, so the usable and effective bases must be stated. A buffer that passes the subtraction method but fails this check is sized to the wrong number.
When the plant has several chillers, the buffer target is set by the smallest active stage, not the plant total. At minimum load only one machine is likely to be running, and that machine's evaporator minimum flow and minimum system volume govern the circuit it shares. Sizing the buffer on the sum of all chillers sizes it to a condition that never occurs.
The practical rule is to size for the single smallest chiller that can be the sole operating unit under the design part-load profile, then confirm the result against the run-time check at the load that unit sees. If the sequence of operation can run a larger machine alone at low load, its minimum applies instead. The point is that the binding case is the machine with the highest minimum volume per unit of load it can serve, and the documentation should name which unit set the target.
The buffer belongs on the common pump suction header, between the chillers and the distribution pumps, so that the chiller evaporator loop and the IT load loop are decoupled. In this arrangement the chiller can circulate its own minimum flow through the buffer while the load side draws a different flow, and the two never have to match. That decoupling is what stops the chiller from hunting.
Placing the tank on the pressure side or directly on the chiller branch defeats the purpose, because the evaporator then sees the load flow directly. The vessel should also sit where its static height helps the system, with the connections arranged so the tank fills and drains through its full height rather than short-circuiting across a near pair of nozzles. For a buffer this mixing tolerance is wider than for a stratified thermal store, but the connection geometry still decides whether the effective volume equals the nominal volume.
The pipework to the buffer is sized for the transient, not the steady maximum flow. When a chiller trips or a load step arrives, flow can reverse and redistribute through the buffer in seconds, and a pipe sized only on the design maximum can throttle that event or provoke water hammer. Sizing should allow the instantaneous flow the buffer must pass during the worst credible step.
A practical approach sets the connecting pipe diameter from the velocity limit at the transient flow, typically a modest figure such as 1.5 to 2.0 m/s, checked against the peak reverse flow the control sequence can produce rather than the running flow. The pipe and its supports also need to accommodate the pressure excursion the transient creates. Where the loop is large, splitting the connection into two nozzles reduces the velocity each sees and smooths the event.
Connection details decide whether the tank performs to its calculated volume. Separate inlet and outlet nozzles, an air vent at the high point, a drain at the low point, and thermometer wells or sensors at both nozzles let the tank be commissioned and watched. A tank piped with a single combined connection, or without a vent, loses effective volume to air locking or thermal short-circuiting.
· Use separate inlet and outlet nozzles placed to draw through the full height of the tank rather than across a short path.
· Fit a high-point automatic air vent and a low-point drain so the tank can be purged and emptied for inspection.
· Provide temperature sensing at both nozzles so the effective delta-T across the buffer can be measured during commissioning.
· Insulate the vessel and its connections for chilled-water duty to avoid condensation and heat gain, which would otherwise erode the usable volume.
The calculation the reviewer wants is short and self-contained. State V_min_sys and its source, state V_loop and its source, show the subtraction, show the run-time check at Q_min with the effective volume used, and name the chiller that set the target. With those lines, another engineer can reproduce the number from the same documents.
· Chiller minimum system volume and the data sheet reference it came from.
· Loop volume from the hydraulic model and coil schedules, with the basis noted.
· Buffer volume by subtraction, with the nominal, usable and effective volumes distinguished.
· Run-time check result at minimum load, compared against the chiller minimum run and off times.
· Placement statement: tank on common pump suction, decoupling chiller from load.
· Pipe sizing basis: transient flow and velocity limit used, not the steady maximum.
Variable | Symbol | How It Is Set | Read From |
Chiller minimum system volume | V_min_sys | Required circuit volume so the evaporator stays in flow limits at part load | Chiller data sheet / AHRI application guidance |
Existing loop volume | V_loop | Piping plus coil, heat-exchanger and equipment inventory | Hydraulic model and coil schedules |
Required buffer volume | V_buffer | V_min_sys minus V_loop; built only if positive | Worked subtraction calculation |
Minimum cooling load | Q_min | Cooling demand at the lowest credible part-load condition | Part-load profile / design basis |
Allowable run time | t | Short-cycle limit the chiller control requires | Chiller minimum run and off times |
Design temperature difference | dT | Selected loop delta-T between supply and return | System design / AHRI rating |
Item | Requirement | Reason | Failure Mode If Omitted |
Tank location | On common pump suction header | Decouples chiller evaporator loop from IT load loop | Chiller short-cycles on load flow |
Inlet / outlet | Separate nozzles through full height | Maximises effective volume | Thermal short-circuit across tank |
Pipe sizing | Sized for transient flow, 1.5-2.0 m/s | Captures flow reversal without hammer | Water hammer or throttled event |
Air vent | Automatic at high point | Purges air so volume stays usable | Air-locked, reduced effective volume |
Drain | At low point | Allows emptying for inspection | Tank cannot be serviced in place |
Insulation | Chilled-water-duty insulation | Prevents condensation and heat gain | Usable volume eroded by gain |
Sensing | Temperature at both nozzles | Lets effective delta-T be measured | Performance never verified |
Center Enamel (Shijiazhuang Zhengzhong Technology Co., Ltd.) designs and fabricates bolted steel storage tanks, welded steel tanks, fusion-bonded epoxy tanks, galvanized tanks, stainless steel vessels, pressure vessels, aluminum geodesic dome roofs and floating roofs. The company has more than 30 years of production experience and has supplied equipment to projects in more than 100 countries. Aluminum geodesic dome roofs are designed and fabricated to API 650 Appendix G, AWWA D108, the Aluminum Design Manual, ASCE 7 and IBC, and the company developed and exported a 72 m diameter aluminum dome roof in 2023. Engineering, fabrication, coating, quality control and site service are handled as one scope, which is why the drawings, the calculation basis and the delivered hardware can be reconciled against each other.
How do you size a chilled water buffer tank for a data centre?
Size the buffer by subtraction. Read the chiller minimum system volume from the chiller data sheet, subtract the volume the loop already holds from piping, coils and equipment, and the remainder is the buffer volume. Then run a run-time check at minimum load to confirm the effective volume holds temperature within band long enough to keep the chiller out of short-cycling. A buffer sized on plant total or on a litres-per-kilowatt rule of thumb answers the wrong question.
What is the rule of thumb for buffer tank volume?
There is no single safe rule of thumb, because the governing number is the chiller's own minimum system volume, which varies by machine and is read from the data sheet. A generic litres-per-tonne figure can be used only as a sanity check after the subtraction method, never as the basis. The defensible volume is V_min_sys minus the actual loop volume, confirmed by the run-time check at minimum load.
Should the buffer tank be sized on the total plant or the smallest chiller?
Size on the smallest active chiller stage, not the plant total. At minimum load one machine is usually the sole operating unit, and that unit's evaporator minimum flow and minimum system volume govern the shared circuit. Sizing on the sum of all chillers sizes the tank to a condition that never occurs. Document which unit set the target and confirm the run-time check against the load that unit actually sees.
Where should the buffer tank be installed?
Install the buffer on the common pump suction header, between the chillers and the distribution pumps, so the chiller evaporator loop and the IT load loop are hydraulically decoupled. That lets the chiller circulate its own minimum flow through the tank while the load draws a different flow. Placing the tank on the pressure side or on the chiller branch defeats the decoupling and lets the evaporator see the load flow directly.
Why does the connecting pipe need transient sizing?
The buffer creates transients: when a chiller trips or a load step arrives, flow can reverse and redistribute through the tank in seconds. A pipe sized only on steady maximum flow can throttle that event or cause water hammer. Size the connecting pipe from the velocity limit at the worst credible transient flow the control sequence can produce, typically 1.5 to 2.0 m/s, and support it for the pressure excursion.
How does delta-T affect the required volume?
Delta-T sets how much temperature drift a given volume buys. In the sensible heat balance t = rho c V_eff dT / Q, a larger allowable temperature swing dT or a larger effective volume V_eff both extend the hold time at minimum load. If you design the loop for a wider delta-T, you need less buffer volume for the same run time; a narrow delta-T demands more. The effect runs through the run-time check, so the two are sized together, not separately.
Do you need a buffer tank if the loop already has volume?
Only if the loop already meets the chiller minimum system volume. The subtraction method gives a buffer volume only when V_min_sys exceeds V_loop; a negative result means the existing piping and coils satisfy the chiller and no tank is required. Do not add a buffer by habit. Confirm the actual loop volume from the hydraulic model first, because a large distributed coil inventory can already provide the decoupling the chiller needs.
A chilled water buffer tank for a data center is sized by subtraction and confirmed by a run-time check, not by a rule of thumb. Take the chiller's minimum system volume from the data sheet, subtract what the loop already holds, size for the smallest active chiller stage, and prove the effective volume holds temperature within band at minimum load.
Place the vessel on the common pump suction so the chiller and the IT load are decoupled, and size the connecting pipework for the transient the tank creates rather than the steady maximum flow. Send us the chiller data sheets, the loop volume from your hydraulic model and the part-load profile, and we will return a buffer volume with the subtraction, the run-time check and the pipe sizing written out for the reviewer.
Send the duty, the volume basis, the governing code and the site constraints, and our engineering team will come back with a specification that a supplier can price and a reviewer can approve. We will also tell you plainly when a different tank type or a different scope is the better answer.