
A buffer tank protects a data center chilled water loop against transients by providing thermal inertia during the chiller's response lag. Size it from the allowable supply temperature excursion and the chiller response time, V = Q t / (rho c dT), then verify the peak excursion and settling time at commissioning with an injected load step.
A chilled water buffer tank in a data center protects the critical load against transients, the short-lived departures from steady state that a chiller cannot answer instantly. This is a chilled-water, HVAC-duty vessel, not a fire-water storage tank, and the NFPA 22 fire-protection requirements that govern fire-water duty do not apply to it. The references that matter here are the chiller response characteristic, the ASHRAE design guidance and a control sequence such as ASHRAE Guideline 36 that defines how the plant reacts to a load change.
A transient is a change in flow or temperature with time, and in a chilled water loop it is unavoidable: chillers stage on and off, pumps change speed, and the IT load steps. The chiller has a finite response, and during that lag the supply temperature drifts. The buffer is the thermal inertia that absorbs the mismatch so the drift stays inside the allowable band. This article sizes that protection against the excursion itself rather than against an arbitrary volume, and shows how to prove it at commissioning.
Throughout, volume means the effective volume the chiller can actually use, not the nominal tank volume. A buffer that is the right size on paper but thermally short-circuited in its connections gives less protection than its nameplate suggests, which is why placement and commissioning are treated as part of the sizing, not as afterthoughts.
A transient is any departure of flow or temperature from the steady-state design condition, measured against time. In a chilled water loop the steady state is the design supply and return temperature at the design flow; a transient is what happens when either moves because something changed upstream or downstream. The loop is never truly steady in a data center, because the IT load, the chillers and the pumps are all adjusting continuously.
The distinction that matters for sizing is between a step and a ramp. A step is a near-instantaneous change, such as a chiller tripping or a large load appearing, and it is the hardest event for the plant to absorb. A ramp is a gradual change the chiller can track. A buffer is sized for the step, because that is the event that produces the largest short-term mismatch between load and chiller output.
When the IT load steps up, the return water temperature rises and that warmer water travels around the loop at the flow velocity, reaching the chiller evaporator after a transport delay set by the pipe volume and the flow rate. The chiller does not see the load change until the warmed water arrives, and then it cannot fully answer it immediately: it has a control deadband, a staging delay, a ramp rate and a minimum run time.
During that response lag the cooling delivered lags the cooling demanded, and the shortfall shows up as a drift in supply temperature. The supply temperature is what the critical load actually receives, so the drift at the supply sensor is the number the design must keep inside its band. A loop without buffering passes the full lag straight to the IT cooling, which is why a single chiller trip can push supply temperature past the limit before any recovery.
The buffer's role in the transient is to be the thermal flywheel. Because it holds a large mass of water at the design temperature, it supplies cooling from its own inventory during the chiller response lag, so the supply temperature drifts by less than it would with no buffer. The larger the effective volume relative to the shortfall, the smaller and slower the excursion.
This is why the buffer is placed to decouple the chiller from the load: it lets the chiller keep circulating its minimum flow and recovering while the load draws from the tank. The protection is not free capacity against a steady deficit; it is capacity available for the seconds to minutes of the response lag. A buffer sized for the transient is therefore sized on time and temperature, not on total plant volume.
The allowable supply temperature excursion is the maximum rise above the supply setpoint the critical cooling can tolerate before the design margin is lost. It is set by the data hall cooling design, the recommended supply temperatures in the ASHRAE guidance, and the control tolerance the owner accepts, not by the tank. A tight setpoint with a small allowable excursion demands more buffer than a loop with margin built into the coil selection.
Stating the allowable excursion explicitly is what turns buffer sizing from a guess into a calculation. The designer names dT_allow, the supply temperature rise the system can absorb, and the rest of the sizing follows from it. Without that number the buffer can only be specified by rule of thumb, and a rule of thumb cannot demonstrate that the critical load stays protected through the worst credible step.
Sizing against the excursion rather than the volume starts from the sensible heat balance over the chiller response time t:
1. V_eff = Q_peak * t_response / (rho * c * dT_allow), where Q_peak is the cooling shortfall during the lag, t_response is the chiller response time, and dT_allow is the allowable supply temperature excursion.
2. Q_peak is the difference between the load step and what the chiller can deliver during t_response, taken from the chiller ramp and staging data.
3. Check the result against the decoupling volume the chiller minimum system volume requires, and use the larger of the two as the effective volume.
The method makes the volume a consequence of three defended numbers: the peak shortfall, the response time and the allowable excursion. It also shows the trade-offs directly: a longer tolerated excursion, a faster chiller response or a smaller peak shortfall all reduce the required tank, which is information a reviewer can act on.
Transient performance is verified at commissioning by injecting a known load step and recording what the supply temperature does. The test applies a step the design assumed, logs supply and return temperature and flow through the event, and measures the peak excursion and the settling time against the band the design stated.
· Inject the design load step, or the largest credible step the sequence can produce, with the chiller in its worst-case starting condition.
· Log supply temperature at the critical sensor and the flow, at a sample rate fine enough to catch the peak.
· Record the peak excursion and the time to return within the band.
· Compare both against dT_allow and the settling time the design required; the buffer passes only if both are met.
· Repeat after any change to the control sequence, because the response time the sizing assumed lives in the sequence.
Not every transient comes from the IT load. Pump start and stop events, chiller sequencing and the deadband around each staged unit, bypass-valve modulation and the opening of a makeup or equalizing line all move flow and temperature in the loop. On the heat-rejection side, a condenser or cooling-tower transient couples through the chiller and appears at the supply a moment later.
These non-IT transients are worth listing in the design because the buffer must absorb the largest credible one, whatever its source. A commissioning test that only exercises the IT load step can pass while the plant still sees an unacceptable excursion from a chiller staging event. The control sequence, the pump arrangement and the bypass logic are therefore part of the transient picture, not separate from it.
The buffer approach has limits that the design must state, because overselling it leads to a tank that costs volume and money without solving the problem. A buffer absorbs transients; it does not fix a plant that is undersized for the steady demand, and it does not replace low-temperature water if the setpoint itself is wrong. It also cannot help if it is thermally short-circuited by poor connections, because then its effective volume is far below its nominal volume.
There is a cost and a footprint to every cubic metre of tank, and beyond a certain excursion the answer is a faster chiller response or a revised sequence rather than more steel. The honest statement for the reviewer is that the buffer is one term in the transient budget, sized against a named allowable excursion and verified, not a catch-all for every cooling deficiency in the plant.
Event | Source | Effect On Supply | What The Buffer Does |
IT load step up | Data hall demand change | Return warms, supply drifts during chiller lag | Supplies cooling from inventory |
Chiller trip | Failure or staging | Loss of capacity, supply rise | Holds temperature through restart |
Pump start / stop | Hydraulic event | Flow transient, temperature swing | Damps the flow-driven swing |
Chiller staging deadband | Control sequence | Repeated small steps | Smooths the step train |
Bypass valve modulation | Flow control | Return temperature disturbance | Buffers the disturbance |
Condenser / tower transient | Heat rejection side | Couples through chiller to supply | Absorbs the coupled lag |
Step | Action | Record | Pass Criterion |
Setup | Chiller in worst-case starting condition | Initial supply temperature and flow | At design setpoint |
Inject | Apply design load step | Step magnitude and time | Equals assumed event |
Log | Sample supply temperature and flow finely | Time series through event | Peak captured |
Measure | Read peak excursion and settling time | Peak dT, t_settle | Within dT_allow and design time |
Compare | Check against band | Delta vs design | Both met to pass |
Repeat | After any sequence change | Updated results | Still within band |
No verified project record in the published case library matches a chilled water buffer tank specified for transient protection of a data center, so no completed installation is presented here. The auditable content is the selection framework and the calculation the article carries: the allowable supply temperature excursion, the chiller response time, the peak shortfall and the resulting effective volume, with the commissioning measurement method. A specifier should confirm the chiller response characteristic and the allowable excursion from the project design before adopting any volume stated here.
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.
What transient does a buffer tank protect against?
A buffer tank protects against the short-lived supply-temperature excursion that follows a load step or a chiller event, during the interval before the chiller responds. When the IT load steps or a chiller trips, the chiller cannot answer instantly because of its staging delay, ramp rate and minimum run time, and the supply temperature drifts. The buffer supplies cooling from its own inventory during that lag so the drift stays inside the allowable band. It is sized for the step, the hardest event to absorb.
How long can supply temperature drift before it matters?
There is no single time, because it depends on the cooling margin the data hall design carries and the allowable supply temperature excursion the owner accepts. A loop with margin built into the coil selection tolerates a longer or larger drift than a loop running at a tight setpoint. The design states the allowable excursion explicitly; the buffer is then sized so the peak drift at the supply sensor stays within that number through the worst credible step, whatever the clock time.
Is buffer sizing based on volume or on temperature excursion?
Sizing is based on the temperature excursion, not on volume as an independent target. The volume follows from V_eff = Q_peak t_response / (rho c dT_allow), where the peak shortfall, the chiller response time and the allowable excursion set the number. The buffer is then checked against the decoupling volume the chiller minimum system volume requires, and the larger of the two is used. Volume is the output of the calculation, not its starting point.
What causes a transient that is not IT load?
Transients also come from pump start and stop events, chiller sequencing and the deadband around each staged unit, bypass-valve modulation, and the opening of a makeup or equalizing line. On the heat-rejection side, a condenser or cooling-tower transient couples through the chiller and appears at the supply. The buffer must absorb the largest credible event from any source, so commissioning should exercise chiller staging and pump events, not only an IT load step.
How is transient performance verified at commissioning?
Verify it by injecting the design load step with the chiller in its worst-case starting condition, then logging supply temperature and flow through the event at a fine sample rate. Measure the peak excursion and the settling time and compare both against the allowable excursion and the time the design required. Repeat after any change to the control sequence, because the response time the sizing assumed lives in that sequence, not in the tank.
Does a buffer tank remove the need for low-temperature water?
No. A buffer absorbs the transient during the chiller response lag; it does not change the steady-state supply temperature the plant is designed to deliver. If the setpoint or the coil selection is wrong, low-temperature water or a different plant configuration is the correction, and a buffer cannot substitute for it. The buffer is one term in the transient budget, sized against a named allowable excursion, not a remedy for an undersized or mis-set plant.
A chilled water buffer tank protects a data center against transients by acting as thermal inertia during the chiller's response lag, and it is sized against the excursion itself rather than against an arbitrary volume. The calculation is V_eff = Q_peak t_response / (rho c dT_allow), with the allowable supply temperature excursion and the chiller response time stated up front.
The protection is only as good as the commissioning that proves it, so inject the design load step, measure the peak excursion and the settling time, and confirm both against the band the design set. Send us the chiller response data, the allowable excursion and the control sequence, and we will return a buffer volume sized and verified for the transients your plant actually sees.
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.