Chilled Water Buffer Tanks for Data Centers: Sizing, Placement and Loop Design
Created on Today
Chilled Water Buffer Tanks for Data Centers: Sizing, Placement and Loop Design
A chilled water buffer tank in a data center is loop equipment, not a reservoir. Its volume only produces a result when the piping around it is designed for the transient it is meant to absorb, which is why the tank volume and the branch pipe size have to be issued together rather than as separate packages. Two quite different jobs are asked of the vessel. The first is compressor protection: giving the chiller a minimum run time so it does not short-cycle at low part load, with the volume set by minimum load, minimum run time and the loop delta-T. The second is transient ride-through: absorbing a step change in load so the supply temperature stays inside the band the IT equipment tolerates, with the volume set by the transient magnitude and the allowable temperature deviation. The two cases produce different volumes from the same plant, and a tank sized for one does nothing for the other. Gallons per ton is a useful order-of-magnitude check, but it hides both governing quantities and it carries an implied run time and delta-T that the design does not necessarily use, as detailed for how to size a buffer tank for data center.
Two jobs are asked of the same vessel
A buffer tank can be specified for compressor protection, for transient ride-through, or for both, and the three cases do not produce the same vessel. The first question in any sizing is not how big the tank should be but which governing case it is bought for, because the two are solved by different inputs and fail differently, as detailed for how to size a buffer tank for data center.
A tank sized for ride-through can leave the chiller cycling exactly as before, since cycling is a function of run time and load fraction rather than of attached water volume. A tank sized for compressor protection can leave the supply temperature swinging just as far during a load step, because the volume was spent holding the run time, as detailed for how to size a buffer tank for data center.
Aspect | Case A: compressor protection | Case B: transient ride-through |
Governing question | Minimum run time at minimum stable load | Allowable supply temperature deviation |
Primary inputs | Minimum load, run time, loop delta-T | Transient magnitude, duration, deviation |
Symptom if the volume is wrong | Compressor keeps cycling | Excursion still reaches the load |
What this case does not fix | A load peak beyond plant capacity | Short cycling at steady part load |
Case one: compressor protection and the minimum run time
The standard way to protect a compressor is to guarantee a minimum run time at its minimum stable load. The volume that makes this possible is the water needed to absorb, at the design delta-T, the heat rejected in that run time: V = Q x t x 60 / (delta-T x c), with Q the load in kW and t the run time in minutes, as detailed for how to size a buffer tank for data center.
The worked figure is illustrative and shows the method only. A minimum stable load of 600 kW, a 10 minute run time, a 5 K loop delta-T and water at c = 4.186 kJ/kg.K give about 17.2 m3. These are not project figures. A real calculation adds the pipework volume, because the headers move with the tank, as detailed for how to size a buffer tank for data center.
Case two: transient ride-through and the temperature excursion
The second case is a rate-of-change problem rather than a run-time problem. When a block of load steps on, the plant cannot change output instantly, and the difference between load and plant output has to be absorbed by water mass in the loop. The governing quantity is the allowable supply temperature deviation, which comes from the airflow and inlet limits of the load, as detailed for how to size a buffer tank for data center.
Using illustrative inputs only, a load step of 1,200 kW lasting 3 minutes with an allowable mixed-water deviation of 1.5 K requires about 34 m3 of participating water: the transient energy divided by the volumetric heat capacity and the allowable deviation. The result is large because the allowable deviation is small, as detailed for how to size a buffer tank for data center.
The volume calculation, and why gallons per ton is only a check
The calculation is complete only when the participating mass is counted correctly: the tank shell volume the flow reaches, plus the headers and branches that move with it, plus any secondary loop switched in during the event. On small tanks the pipework can be a large fraction of the total, which is why identical volumes behave differently on adjacent loops, as detailed for how to size a buffer tank for data center.
Gallons per ton is an order-of-magnitude check, not a specification, because the number silently embeds a run time and a delta-T. One US gallon per ton holds about 3.8 kg of water, which at 5 K stores roughly 79 kJ per K; at 3.517 kW that is about 2.7 K of movement per minute, as detailed for how to size a buffer tank for data center.
Delta-T drives the size, in both directions
Volume varies inversely with the temperature difference, so the relationship that makes a buffer work at a wide delta-T makes it expensive at a narrow one. Halving the design delta-T doubles the volume: on the illustrative Case A inputs, 10 K gives roughly 8.6 m3, 5 K gives 17.2 m3 and 2.5 K gives 34.4 m3, as detailed for how to size a buffer tank for data center.
Low delta-T is exactly the condition under which short cycling occurs, because the plant is at low load fraction, so a design written at the full-load delta-T understates the volume needed at the condition the tank exists to handle. The calculation therefore uses the delta-T at minimum stable load, and only while the contents are mixed: a stratified buffer absorbs through the layer, not the shell volume, as detailed for how to size a buffer tank for data center.
Placement on the common header, on the pump suction side
The tank goes on the common suction header, so that it is connected to the loop whenever any pump is running and the volume it contributes reaches the plant inlet. A tank on the discharge side cannot smooth the flow the pumps deliver, and a single-pump branch tank works only when that pump is in service, as detailed for how to size a buffer tank for data center.
Two further conditions belong on the same drawing. The water level has to sit above the pump centreline, otherwise the tank drains through the pump and the volume is lost to suction lift. And the connection must let return water enter the upper tank while supply is drawn from the lower cold layer, as detailed for how to size a buffer tank for data center.
Branch pipe sizing at transient flow, dead legs and isolation
The branch connecting the tank to the header is sized on the transient flow the tank has to pass, not the steady distribution flow it sees most of the time. A branch sized on steady flow cannot pass the transient the tank was calculated for, so the volume on the drawing is decorative. The check covers velocity, pressure drop and surge, as detailed for how to size a buffer tank for data center.
Dead legs come from the same geometry. Water that cannot move stagnates, and in a chilled water loop that carries a microbiological risk as well as a freeze risk if the system is drained for winter protection. Isolation valves must not create a pocket against the tank, and low points need a drain. A bypass must carry the full loop flow, as detailed for how to size a buffer tank for data center.
Check | Design input | Consequence if omitted |
Branch size at transient flow | Flow during the load step | Tank cannot absorb the transient |
Velocity limit in the branch | Project velocity limit | Noise and erosion at the connection |
Pressure drop at the connection | Friction loss in the branch | Transient arrives below design rate |
Surge on fill and drain | Fill and drain rates | Water hammer and nozzle damage |
Dead leg at the connection | Low points between isolation valves | Stagnant water; freeze damage |
Bypass and isolation | Full loop flow rating, drain, spool | A shell cannot be removed without draining |
Multiple shells on one header, and the pressure boundary
Where the volume needs more than one shell, the shells sit on the same header and each is individually isolatable, so a shell can be drained and removed while the rest of the system stays in service. Smaller shells also reduce the lifting problem, and mixed sizes are acceptable if sized on the worst single-shell flow, as detailed for how to size a buffer tank for data center.
A verified delivery of eleven epoxy-coated tanks of mixed sizes on a shared header illustrates that arrangement; its medium is chemical wastewater, so it is cited for the arrangement and the isolation logic only. On a closed loop the atmospheric buffer is not available, because the system needs a pressure boundary: the vessel becomes a pressurised item whose relation to the expansion vessel and relief device must be resolved, as detailed for how to size a buffer tank for data center.
Commissioning and the specification that carries the governing case
A buffer tank is the easiest item in a chilled water plant to accept on paper and the easiest to leave unproven in service. The commissioning test has to show the two properties the sizing claimed: that the chiller holds its minimum run time, and that supply temperature stays inside the band the load tolerates during a designed event, as detailed for how to size a buffer tank for data center.
The test starts the chiller at minimum stable load, logs compressor starts per hour, then switches between pumps on the common header and confirms the tank stays connected through the changeover. A designed load step is applied and the tank inlet and outlet temperatures are logged, so the response is compared with the calculated deviation, as detailed for how to size a buffer tank for data center.
If the outlet moves further than predicted, the branch is the first thing to examine, because a restriction at the connection produces the same trace as an undersized tank. A submittal becomes checkable when the request states the design basis: the governing case, the minimum stable load and run time or the transient and allowable deviation, the design delta-T, the pipework volume and the branch sizing, as detailed for how to size a buffer tank for data center.
· Name the governing case: compressor protection, ride-through, or both, as detailed for how to size a buffer tank for data center.
· State the minimum stable load and run time, or the transient size and deviation, as detailed for how to size a buffer tank for data center.
· State the design delta-T at the governing load condition, not at full load, as detailed for how to size a buffer tank for data center.
· Include the pipework volume in the mass balance, as detailed for how to size a buffer tank for data center.
· Locate the tank on the common suction header, level above pump centreline, as detailed for how to size a buffer tank for data center.
· Size the branch at transient flow, with velocity limit and surge provision, as detailed for how to size a buffer tank for data center.
· Require no dead legs between tank and header, with drain points identified, as detailed for how to size a buffer tank for data center.
· Record run time, starts per hour and the measured supply temperature deviation, as detailed for how to size a buffer tank for data center.
Project Case Study
Reference delivery for this duty, cited from the verified project library. Match level C. Eleven mixed-size tanks on one site sharing a common header, each independently isolatable. The medium is chemical wastewater rather than chilled water, so the project is used for the parallel arrangement and maintenance isolation logic only, and not for any chilled water thermal performance statement, as detailed for how to size a buffer tank for data center.
Project ID | Country | Product | Capacity | Tank count | Dimensions | Completion | Supply scope |
CH-WW-4771 | China | Epoxy Coated Tank (FBE) | 4,771 m3 | 11 | Mixed sizes: φ4.59 × 9.6 m (3), φ6.11 × 9.6 m (4) and 4 further tanks of mixed size | 2025-05 | Not recorded |
Eleven mixed-size tanks on one site sharing a common header, each independently isolatable. The medium is chemical wastewater rather than chilled water, so the project is used for the parallel arrangement and maintenance isolation logic only, and not for any chilled water thermal performance statement, as detailed for how to size a buffer tank for data center.
Frequently Asked Questions
Q1. What problem does a buffer tank actually solve?
It solves a hydraulic problem in the loop, not a storage problem. A buffer on the common suction header uses the thermal inertia of the water to hold plant output steady against a changing load. It protects the compressor by guaranteeing a minimum run time, or carries a load step through a period the plant cannot follow, as detailed for how to size a buffer tank for data center.
Q2. How do compressor protection and transient ride-through differ?
Compressor protection is a run-time problem, solved by volume that lets the machine run for a stated minimum period at its minimum stable load. Transient ride-through is a rate-of-change problem, solved by volume that absorbs a load step while mixed-water temperature stays within the deviation the load tolerates. A volume solving one leaves the other unchanged, as detailed for how to size a buffer tank for data center.
Q3. How is the volume calculated and why is a rule of thumb not enough?
The calculation is V = Q x t x 60 / (delta-T x c) for compressor protection, with Q the minimum stable load, t the minimum run time, delta-T the loop difference at minimum load and c the specific heat, plus the participating pipework volume, as detailed for how to size a buffer tank for data center.
Q4. Why does delta-T drive the size?
Heat stored in water is proportional to the temperature difference it can absorb, so the required volume varies inversely with delta-T. Halving the design delta-T doubles the volume: on the illustrative inputs, a 600 kW load with a 10 minute run time gives about 8.6 m3 at 10 K, 17.2 m3 at 5 K and 34.4 m3 at 2.5 K, as detailed for how to size a buffer tank for data center.
Q5. Where must the buffer be installed?
On the common suction header, so it is connected to the loop whenever any pump runs, with the water level above the pump centreline so the tank does not drain through the pump. Return water should enter the upper tank and supply be drawn from the lower cold layer. A single-pump branch tank works only in part of the operating pattern, as detailed for how to size a buffer tank for data center.
Q6. Why must the branch pipe be sized at transient flow?
Because the tank's duty is to pass a transient, and the steady distribution flow it sees most of the time is much smaller. A branch sized on steady flow cannot pass the transient the tank was calculated for, so the volume on the drawing does not exist in practice. The branch also needs checking for velocity, pressure drop and surge on fill or drain, as detailed for how to size a buffer tank for data center.
Q7. How are multiple shells arranged on one header?
Each shell connects to the common header through its own branch, with isolation so one shell can be drained and removed while the remainder stays in service, and with a drain point at each connection. Smaller shells reduce the lifting problem and allow mixed sizes, but the header and branches are sized on the worst single-shell flow, as detailed for how to size a buffer tank for data center.
Q8. How are the pressure boundary and relief considered?
On a closed loop the tank cannot be an open vessel, so the buffer becomes a sealed item and its relationship with the expansion vessel, the pump suction and the relief device must be resolved. The buffer must not be credited as the relief device, because the relief setting has to sit above the allowable pressure of the weakest component, as detailed for how to size a buffer tank for data center.
Q9. How does commissioning prove the sizing was right?
By logging measured values rather than recording a pass. The chiller is started at minimum stable load and compressor starts per hour are recorded against the required run time; the pumps are switched on the common header to confirm the tank stays connected; and a designed load step is applied with inlet and outlet temperatures logged for comparison with the calculated deviation, as detailed for how to size a buffer tank for data center.
A chilled water buffer tank delivers nothing on its own. It delivers when the governing case has been named, the volume calculated from that case rather than from a rule of thumb, the tank connected to the common suction header above the pump centreline, and the branch pipe sized to pass the transient the calculation assumed. The two failure modes are predictable: a tank sized for run-time protection that leaves the compressor cycling as before, and a tank sized for temperature ride-through that leaves the supply temperature swinging as before. Both are avoided by the same discipline: state the governing case, the design delta-T at the governing condition and the branch size, then prove the result in commissioning with measured run time and temperature deviation, as detailed for how to size a buffer tank for data center.
Talk to an Engineer
For a chilled water buffer tank, send the plant arrangement and the loop data: chiller capacity and part-load or turndown data, the minimum run time required by the manufacturer, the loop design delta-T at minimum load, the pipework volume on the common header, the maximum allowable supply temperature deviation, the pump elevations and header topology, and the required supply scope. We will return a sizing study showing the governing case, the calculated volume including the participating pipework, the branch sizing at transient flow with velocity and surge checks, the placement and isolation arrangement, and a commissioning test with recorded acceptance criteria, as detailed for how to size a buffer tank for data center.
About Center Enamel
Shijiazhuang Zhengzhong Technology Co., Ltd (Center Enamel) is a professional manufacturer dedicated to the design and fabrication of bolted storage tanks since 2008. The product range includes Glass-Fused-to-Steel (GFS) tanks, Fusion-Bonded Epoxy (FBE) tanks, stainless steel tanks, galvanized steel tanks and aluminum geodesic dome roofs, together with EPC technical support and tank ancillaries. Production runs under ISO 9001, ISO 14001 and ISO 45001, and the product range is certified to NSF/ANSI 61, WRAS, CE and EN 1090, ISO 28765, AWWA D103-09 and NFPA where the application requires it. The company holds close to 200 enameling patents, runs a 150,000 m2 production base, produces about 300,000 enamel-coated steel plates a year, has completed more than 30,000 installed projects and has shipped tanks to more than 100 countries. The engineering design capability covers tank volumes up to 60,000 m3, as detailed for how to size a buffer tank for data center.
WhatsApp