Data centre noise: why the absolute limit rarely stops it

It is two in the morning, and the acoustics technician stands in the garden of a house two hundred metres from the data centre fence, meter in hand, facing a bedroom window left ajar. At the boundary the reading is already done: noise from the chillers and dry coolers sits comfortably inside the zone limits, generous for an industrial area. That is not where the problem lies. It is the difference between what the meter reads now, plant running, and what it reads in an hour, once the plant stops for a maintenance test. A few decibels decide the case, and the reading that settles a dispute is never the one taken at the fence.

The framework law, and the decree that sets the numbers

The framework is Law No. 447 of 26 October 1995, Italy’s framework law on noise pollution: principles and delegations, not thresholds. Article 3, paragraph 1, letter a) assigns to the State, by DPCM, «dei valori di cui all’articolo 2» (the values referred to in Article 2) — the definitions Article 2 lists without figures. The decree that writes those figures is the DPCM of 14 November 1997, «Determinazione dei valori limite delle sorgenti sonore» (Determination of the limit values of sound sources, Official Gazette No. 280 of 1 December 1997), whose Article 1 says so itself: adopted «in attuazione dell’art. 3, comma 1, lettera a), della legge 26 ottobre 1995, n. 447» (implementing Article 3, paragraph 1, letter a), of Law 447/1995). Two instruments, two jobs: the law says what gets measured, the DPCM says how much is too much.

Three quantities never to confuse

Article 2, paragraph 3, of Law 447/1995 splits immission limits into two families: «valori limite assoluti, determinati con riferimento al livello equivalente di rumore ambientale» (absolute values: referred to the equivalent level of ambient noise), and «valori limite differenziali, determinati con riferimento alla differenza tra il livello equivalente di rumore ambientale ed il rumore residuo» (differential values: the difference between ambient noise and residual noise). A third quantity, emission, is tied to the single source rather than the environment. The 1997 DPCM treats all three in separate articles, each with its own table: emission (Art. 2), measured near the source, Table B; absolute immission (Art. 3), all sources combined, Table C; differential (Art. 4), not a value but a difference, measured inside the dwelling.

The tables, Leq in dB(A), daytime (06.00-22.00) and night-time (22.00-06.00): Table B (emission) — I 45/35 · II 50/40 · III 55/45 · IV 60/50 · V 65/55 · VI 65/65. Table C (immission) — I 50/40 · II 55/45 · III 60/50 · IV 65/55 · V 70/70 · VI 70/70. A specification citing the zone limits without saying which quantity it means has bought nothing: each carries its own thresholds and measurement point.

Article 4: the core of the decree

It is Article 4 of the DPCM that sets the differential limit, the point on which a dispute is won or lost: «I valori limite differenziali di immissione […] sono: 5 dB per il periodo diurno e 3 dB per il periodo notturno, all’interno degli ambienti abitativi. Tali valori non si applicano nelle aree classificate nella classe VI della tabella A allegata al presente decreto» (differential immission limits: 5 dB by day, 3 dB by night, inside residential premises; not applicable in Class VI areas under Table A). Three things matter more than the figures. First: measured inside the dwelling, not at the fence, not on the façade. Second: the night period spans eight hours, 22.00 to 06.00 — not the four or five peak hours often pictured. Third: the night threshold, 3 dB, is tighter than the daytime one, 5 dB, exactly when the chillers keep running. The Class VI exclusion is no loophole: by Table A’s own definition those areas are «prive di insediamenti abitativi» (free of residential settlement), so there is no receptor left to protect.

The threshold that switches the differential off

The differential is not an absolute principle: it rests on numeric thresholds, and below them Article 4 states that «ogni effetto del rumore è da ritenersi trascurabile» (any noise effect is deemed negligible), and the rule does not apply at all. Two alternative cases: «se il rumore misurato a finestre aperte sia inferiore a 50 dB(A) durante il periodo diurno e 40 dB(A) durante il periodo notturno» (windows open: below 50 dB(A) by day, 40 dB(A) by night), or «se il livello del rumore ambientale misurato a finestre chiuse sia inferiore a 35 dB(A) […] e 25 dB(A) […]» (windows closed: below 35 dB(A) by day, 25 dB(A) by night). The decree also excludes noise from road, rail, airport and maritime infrastructure, behaviour unconnected with productive needs, and, verbatim, that produced «da servizi e impianti fissi dell’edificio adibiti ad uso comune, limitatamente al disturbo provocato all’interno dello stesso» (fixed building services and plant for shared use, limited to disturbance inside that same building). Whoever drafts the specification and whoever disputes it argue over where the microphone stands and at which window: windows closed, below 25 dB(A) at night — a quiet bedroom in the countryside — and the differential does not apply.

Why it breaks at night, not by day

A data centre is a fixed, continuous source: chillers and dry coolers run even as the IT load drops. The site almost always sits in Class V or IV; the nearest homes almost never do. On the absolute limit there is headroom — up to 70 dB(A) even at night in Class V. It is the differential that tightens: the residual noise — what would remain with the plant off — in a residential area at night is very low, often well under 35 dB(A). A handful of decibels of fan noise, inside a bedroom with the window shut, is enough to exceed the 3 dB night threshold, precisely when the plant keeps running and the background drops furthest. The differential is measured where the rule says: inside the dwelling, not at the fence.

What the specification has to say before the plant is switched on

An acoustic impact report or a specification meant to hold up once the plant runs states five things: the quantity — emission, absolute or differential; the point — at the fence for the absolute limit, inside the most exposed dwelling for the differential, both window states declared; the period — daytime 06.00-22.00 or night-time 22.00-06.00, a genuine eight hours; the instrument, with its accuracy class; and the record of its calibration — an uncalibrated meter produces a number, not a measurement that holds up. Acceptance signed off without these five lines reopens at the first complaint, once January’s residual noise is no longer July’s and no one remembers which instrument took the first reading.

See the service · Talk to an engineer

What we don’t know

Article 2, paragraph 2, of the DPCM explicitly conditions Table B: the values stand «fino all’emanazione della specifica norma UNI che sarà adottata con le stesse procedure del presente decreto» (pending the specific UNI standard, to be adopted via the same procedure as this decree). Almost thirty years on, we have found no primary source — not in Normattiva, not in the UNI catalogue, not in an official update to the decree — establishing whether that standard was ever issued through the procedure the decree requires, a DPCM adopting it. We take no position either way: we do not know, and we say so rather than assume it.

Two threads, applied to this account

The first thread is stating, for every measurement, the quantity, point, period and instrument listed above — checking at acceptance that the calibration certificate exists, not that someone says it does. The second thread is what makes that measurement useful beyond the test day: with CSIDIA, the group’s other company, night-time readings, the reference residual noise and the chillers’ load cycles join the same single model on which an AI runs the diagnosis: a differential edging toward the limit in July, when the dry coolers work harder, is not the same finding as one flat all year — a comparison that holds only if every reading carries its own date, point and instrument. Within the client’s own perimeter: on-premise, on autonomous machines with no deep integration into the existing network, or a dedicated cloud with a data centre in Italy, always with shared management between the two teams.

The site visit, at no cost, produces a map of your acoustic measurement points — with, for each one, which quantity is being checked, at which window, over which period, and whether the instrument used is calibrated. Including the boxes that stay empty: that list is yours to keep even if we don’t go on to work together.