Does your data centre’s cooling pass the test? ITU-T L.1322 says how to measure it

The specification for expanding the server room calls for a cooling test. The tester turns up with last quarter’s PUE: it’s within the contractual threshold, the air conditioners are running at the expected load, the average temperatures check out. Everything passes on the report. Then someone asks the awkward question: does the temperature each individual cabinet actually pulls in — not the room average — say the same thing in every aisle? That’s the question a PUE never answers, because it’s a whole-room figure, not a row figure, not a cabinet figure. That is exactly the gap a recommendation published in December 2025, ITU-T L.1322, tries to close: in the worked example in its own appendix — a 20 × 16 metre hall, six rows of cabinets, three cold aisles and two hot aisles, five air conditioning units — the general parameters check out, but the average air-mixing index is about 27% above the threshold for a good result. On the summary sheet, the cooling works. It doesn’t work the same way everywhere in the room.

A December 2025 recommendation, with metrics across five levels

ITU-T L.1322, Multi-level metrics for thermal environment and thermal performance of data centres, approved on 14 December 2025 by Study Group 5, first and only edition, we downloaded and read in full: it is free and does not show as “deleted”. The scope, verbatim: “This Recommendation provides a standardized framework for evaluating the thermal environment and thermal performance of data centres.” The summary spells out the method: “The framework defines multi-level indicators: room-, row-, cabinet-, device-, and chip-level, covering airflow organization and cooling effectiveness.” It does not replace the PUE of ITU-T L.1302, which we have already written about: it sits alongside it, shifting the question from how much energy does cooling consume to does the cold air actually reach where it is needed.

The whole-room number: SHI, RHI, RTI

The first level assesses the entire room. The supply heat index (SHI) “measures the extent to which hot air […] mixes into the cold aisle at the cabinet air inlet.” “A lower SHI thus signifies better airflow organization and a higher-quality thermal environment”: the closer to zero, the better. Its counterpart, the return heat index (RHI), looks at the opposite path — how much cold air ends up in the hot aisle — and in the ideal case “SHI + RHI = 1, representing perfect airflow separation and optimal cooling performance.” The third index, the return temperature index (RTI), tells you which way the system is failing: “An RTI greater than 100% indicates significant hot air recirculation, while an RTI less than 100% indicates pronounced cold air bypass or short-circuiting,” with “RTI = 1” as the design reference value. Three different numbers, because a system can have heat recirculation, cold-air short-circuiting, or both, in different spots of the same room.

From the aisle to the cabinet: where the hot spot hides

The next level down is the row of cabinets: the index of mixing (IOM) and its heat-weighted variant, HIOM, flag which rows are most likely to harbour a hot spot. At the single-cabinet level, the cabinet cooling index (RCI) compares each cabinet’s inlet temperature against the recommended range, and the recommendation sets a four-tier classification: Ideal 100%, Good 96%-100%, Acceptable 90%-96%, Poor ≤90%. The text is unambiguous: “A value of 100% for RCI means an ideal condition; namely, no temperature is too high or too low, and the temperature corresponds to the recommended temperature range.” For high-density cabinets with liquid or hybrid cooling, the recommendation goes further still, down to device and chip level, with indicators such as thermal design power (TDP) and thermal resistance — useful once air alone can no longer carry the heat away.

What the worked appendix example shows

The recommendation’s appendix — not normative, but built to illustrate the method — applies the sequence to a model hall: cabinets dissipating 1 to 13 kW, constant cooling airflow per cabinet, air conditioners with a 20°C supply temperature. One figure explains the problem on its own: “The ratio of the air volume of air-conditioning supply to the air volume of the IT equipment demand is 0.83” — the air conditioners deliver 83% of the airflow the equipment would call for. From there, the average SHI works out at 0.255 against a good-result threshold cited as 0.2 — a figure drawn from a bibliographic reference within the recommendation that we have not verified against its paywalled original source. Dropping to row level, HIOM isolates rows 3 and 4: “Rows 3 and 4 have significantly stronger heat potential (greater than 1) than the other rows, and self-circulation of high-temperature exhaust air may occur.” Dropping to cabinet level, the problem narrows further: “the larger fluctuations on both sides indicate that the cabinets at both ends have larger potential problems” — it is the end-of-row cabinets, not the central ones, that suffer the recirculation. Three separate tests, on the same room, to get from cooling is adequate to it’s these two cabinets, in these two rows.

What to put in the specification

The test level, not just the final figure: room-level SHI and RTI, row-level HIOM, cabinet-level RCI and mixing index for every critical cabinet — not an overall verdict with no detail behind it.
The point and instrument for every temperature reading: inlet, outlet and supply for each cabinet measured, with a calibrated, traceable instrument behind every reading, exactly as for any other acceptance measurement.
An acceptance threshold declared per level, set before the measurement is taken: Good/Acceptable/Poor for RCI, not a bare compliant with no tier attached.
The ratio between delivered and required airflow, the figure that on its own explained the shortfall in the worked example, before hunting for the cause elsewhere.
The link to environmental control under EN 50600-2-3: the availability class declared at design stage and the indices measured in operation need to tell the same story.

The point

A data centre can meet its contractual PUE and still have two rows out of six with hot-air recirculation: these are different measurements, at different levels, and neither substitutes for the other. Compliance with L.1322 remains voluntary, as with any ITU-T recommendation — but a specification that ignores it accepts a test that sees the room and misses the cabinet. That is why, in the acceptance tests we run, the measurement level is a line written before the site visit, not a choice made on the spot. And for a data centre, together with CSIDIA, the group’s other company, per-cabinet temperatures, BMS data and equipment inventory stop being three separate archives and become a single map, on which an AI flags the drift between rows before it turns into a failure — within the client’s own perimeter, on-premise or on a dedicated cloud with a data centre in Italy.

Do you need to test a data centre’s cooling, or check one already handed over? Talk to an engineer: the room-level number reads off a report; the hot spot only turns up once you drop a level.