In a rack or a data hall, the diagnosis "we do not have enough cooling" is usually wrong. The total capacity is sufficient; what is insufficient is the cold air reaching the place that needs it. Two loss mechanisms dominate:
Bypass air. Cold air escapes to the return path without passing through any equipment: open rack slots, unsealed cable cut-outs, leaks in the raised floor.
Recirculation. Hot exhaust air finds its way back to the intake. This is why the top slots of a rack often run ten degrees hotter than the bottom ones.
Neither shows up in a total-capacity calculation; both appear only in a field solution. Data centre and enclosure problems are therefore the classic case where the question is about the distribution, not the mean.
The acceptance criterion is intake temperature, not room temperature
ASHRAE TC 9.9 criteria are defined at the equipment intake, not as a room average [1]. The recommended envelope is 18–27 °C for classes A1–A4; the allowable envelope widens by class, from 15–32 °C for A1 to 5–45 °C for A4.
That is what the analysis should monitor: the temperature distribution over each rack's intake face. "The room averages 24 °C" means nothing while a server in the top slot is drawing air at 38 °C.
On the component side: the thermal resistance chain
At board and component scale the acceptance criterion is junction temperature, and the thermal resistance figures the manufacturer quotes are measured to the JEDEC JESD51 series [2]. The point to watch is that these values are measured on a standard test board in a defined environment — JESD51-2 defines the natural convection, still-air environment [3].
The flow in your actual enclosure is not that environment. Applying a catalogue junction-to-ambient resistance directly to your own design is one of the most common and quietest errors. The right approach is to solve board and enclosure conjugately and use the catalogue figure only as an upper-bound check.
Modelling decisions
Do not model each server. The internal geometry of dozens of servers can neither be resolved nor needs to be. Each unit is represented as a porous block with a known flow rate and heat load. What matters is that the intake and exhaust faces are in the right place.
Give the fan curve; do not impose the flow rate. Server fans are not constant-flow devices; as back pressure rises, flow falls. Assuming a fixed flow rate erases the effect of blockage from the model. See the note on fan operating points.
Do not neglect radiation in sealed enclosures. In a low-density, naturally cooled enclosure, radiation carries a significant share of the total heat rejection [4].
References
- ASHRAE Technical Committee 9.9 (2021). Thermal Guidelines for Data Processing Environments, 5th ed. ASHRAE, Atlanta.
- JEDEC Solid State Technology Association. JESD51 — Methodology for the Thermal Measurement of Component Packages (Single Semiconductor Device).
- JEDEC Solid State Technology Association. JESD51-2 — Integrated Circuits Thermal Test Method Environmental Conditions: Natural Convection (Still Air).
- Modest, M. F. & Mazumder, S. (2021). Radiative Heat Transfer, 4th ed. Academic Press. ISBN 978-0-323-98406-5.