Hot Rack Tops in an Uncontained Hall
A 400 kW hall houses 12 rows of eight racks on a 600 mm raised floor, cooled by four 110 kW chilled water CRAHs with no aisle containment. Average rack load is 4 kW, but rows 5 and 6 contain six storage arrays drawing 11–12 kW each. The BMS shows supply air at 15 °C, return air at 27 °C, and 12–18 Pa under the floor at the far end of the room. Thermal surveys over the past month record inlet temperatures of 19–21 °C at the bottom U positions of most racks but 29–33 °C at the top U positions of the high-density racks in rows 5 and 6. A vendor has proposed lowering the CRAH set point to 12 °C and adding two portable spot coolers as an interim fix. The operations team has already fitted blanking panels in rows 5 and 6, but the cable cut-outs beneath those racks remain open, and six racks were removed last quarter whose floor tiles and perforated grates were never replaced.
1. Interpret the temperature pattern. What does it tell you, and why does the vendor's proposal mask rather than fix the problem?
Read the worked answer
The pattern - normal temperatures at the bottom of racks, 29–33 °C at the top U positions of the 11–12 kW racks, with 15 °C supply air - is the signature of recirculation, not of insufficient supply temperature. Hot exhaust from the dense racks is being drawn back into the rack fronts at the top, where the pressure differential across the rack is weakest. Cold-aisle air never reaches the upper U positions, so the equipment there is cooled by its own exhaust.
The numbers support a second, simultaneous problem: a return air temperature of only 27 °C against 15 °C supply is a 12 K delta-T, lower than you would expect if all the supply air were doing useful work at a 400 kW IT load. That means a significant fraction of the cold air is bypassing the IT load - through open cable cut-outs and the missing tiles - so CRAH capacity is being spent on air that cools nothing. The 12–18 Pa under-floor pressure at the far end is on the low side, which is consistent with leakage near the CRAHs starving the far rows.
Lowering the set point to 12 °C fixes none of this and makes some of it worse: it increases cooling energy, raises condensation risk on uninsulated surfaces, and the larger supply-to-room temperature difference increases buoyancy-driven recirculation because cold air dumps faster through the nearest openings. Portable spot coolers treat one aisle and can disturb the pressure field elsewhere.
There is also a compliance point: 33 °C at the inlet exceeds ASHRAE recommended inlet limits (18–27 °C) and even the allowable upper bound for Class A1 equipment. This is a warranty and reliability issue, not a comfort issue.
2. Prioritise the corrective measures and justify the order.
Read the worked answer
Order by speed, cost and whether the action restores the pressure field the rest of the fixes depend on.
- Seal the floor - first, immediately. Blanking panels alone do not stop the bypass. Fit grommets or brush seals to every cable cut-out under rows 5 and 6, replace missing tiles with solid ones, and remove grates that are not in front of a heat load. This is low cost, can be done in one maintenance window, and restores under-floor static pressure to the far rows. Every later step is more effective once this is done.
- Right-size and balance the perforated tiles. A 20 kW-class rack needs on the order of 4,000–5,000 m³/h at a 12 K rise (P = 1.2 × volumetric flow × ΔT). A standard 600 × 600 mm tile passes only around 800–1,200 m³/h at normal plenum pressures, so the dense racks need multiple high-flow grates or direct ducted supply - and the room needs a commissioning airflow check row by row, not a visual count.
- Contain the high-density rows, then the room. Cold-aisle containment is the durable fix for recirculation, but it must follow steps 1 and 2: containing a leaky aisle simply pressurises the leaks and pushes bypass elsewhere.
- Add rack-level cooling only if needed. If rear-door or in-row heat exchangers are required for the 11–12 kW racks after containment, they need their own cooling water or refrigerant provision and must be included in the redundancy analysis.
- Then re-tune the plant. Raise plenum pressure to the target (aim for the upper end of the pressure guidance at the furthest row), review CRAH fan speed and set points, and only then revisit the 15 °C supply temperature - with containment in place it can usually rise, which saves energy rather than costing it.
3. What measurements would you take to confirm the diagnosis and to accept the remediation, and what acceptance criteria would you set?
Read the worked answer
Diagnosis needs three independent measurements, because each one alone can mislead:
- Rack inlet temperature at top, middle and bottom of every rack, with a logged survey over at least one full day and at peak load - not a single snapshot. The top-of-rack reading is the one that matters and it is the one a spot check usually misses.
- Under-floor static pressure at the head of each row, plus the pressure differential across each CRAH. This shows whether the far rows are being starved and whether leakage is near the plant or distributed.
- Airflow and delta-T per CRAH, and an estimate of the airflow actually reaching the IT load. Comparing CRAH airflow against the IT heat load through P = 1.2 × volumetric flow × ΔT lets you quantify the bypass fraction and proves whether you are buying containment or just more fan energy.
Acceptance criteria to write into the remediation scope:
- No rack inlet exceeds the ASHRAE recommended range at any U position under design load - target 18–27 °C, with the upper bound treated as a hard limit for the dense racks.
- Rack delta-T remains stable at the top U positions during a load step, proving recirculation is controlled.
- Measured CRAH delta-T approaches design at full load, and the bypass fraction drops to a documented low figure.
- Under-floor pressure at the furthest row stays within specification with all containment and sealing in place.
Report the results as a before/after table with the same instruments and the same load, otherwise the comparison proves nothing.