The GPU pilot, the failed CRAH and the setpoint shortcut
A 900 square metre hall holds 40 racks at a design density of 8 kW, a total of 320 kW of IT load. Cooling comes from four chilled-water CRAHs rated 110 kW each, three duty and one standby, with hot-aisle containment and a 24 degrees C cold-aisle setpoint. Last month a GPU pilot added four racks at 12 kW, accepted because 'there was spare capacity'. This week CRAH-3 developed a bearing fault and is isolated for a ten-day parts lead time, leaving three units running continuously. The DCIM shows the highest rack inlet at 31.5 degrees C. The facilities team proposes raising the cold-aisle setpoint to 26 degrees C 'because ASHRAE allows 27 degrees C'. The IT equipment is rated to ASHRAE A2, where recommended inlet temperatures top out at 27 degrees C. The IT director wants a written position before the weekend, because two more 12 kW racks are due on Monday.
1. Is the site currently running N+1? Determine the actual thermal headroom and set out your actions for the ten-day window.
Read the worked answer
No. With CRAH-3 isolated, the hall is running three units with no standby, so it is N at best. The nominal arithmetic is already negative: 3 x 110 kW gives 330 kW of nominal sensible capacity against 368 kW of installed IT after the pilot. Nameplate ratings also flatter the plant, because CRAH capacity falls as chilled-water supply temperature rises and as airflow is disturbed by containment gaps and bypass, which is why one rack inlet is at 31.5 degrees C even if the hall average looks acceptable.
Actions for the window: refuse Monday's two racks, in writing, against the capacity position. Deploy spot cooling on the hot aisle, powered from a monitored circuit with its current draw recorded. Lower the chilled-water supply temperature if the plant has headroom, and rebalance airflow with blanking panels, floor grilles and fan speeds. Increase monitoring frequency to the affected racks and set a defined trip point for shedding the GPU pilot load. Escalate the parts lead time as a supply chain risk and ask whether a temporary rental CRAH can be brought in. Record the residual risk and its acceptor before Friday.
2. Respond to the proposal to raise the cold-aisle setpoint to 26 degrees C.
Read the worked answer
It makes the problem worse while appearing to solve it. Raising the setpoint raises rack inlet temperatures; the 31.5 degrees C rack would go higher, not lower. ASHRAE's recommended envelope for A2 equipment tops out at 27 degrees C, and the allowable envelope, currently 10 to 35 degrees C for A2, is a survivability limit for short-term excursions, not an operating target. Even inside the allowable range, sustained operation at high inlet temperature shortens hardware life and raises component failure rates, and many vendors will query warranty claims made in that zone.
The team is also confusing two different measurements. A cold-aisle or room setpoint of 24 degrees C routinely coexists with a 31.5 degrees C rack inlet because of recirculation, bypass airflow and containment gaps. The fix is airflow management and enough cooling capacity, not a setpoint change. If a setpoint change is made at all, it must be justified by measured rack inlet temperatures with all CRAHs healthy, and it must be documented with the resulting inlet map.
The defensible position for the IT director is: hold or reduce the setpoint during the degraded window, fix the airflow, and revisit setpoint optimisation only after CRAH-3 is back and the hall is measured at design conditions.
3. What capacity governance failure allowed this, and what control would you put in place?
Read the worked answer
The pilot was approved against 'spare capacity' defined as nameplate capacity minus current load, with no check of what redundancy state remained and no check at design ambient. That definition is the root cause. Spare capacity in a data centre means headroom above the load that keeps the designed redundancy intact at the worst design condition, not the gap between today's load and the sum of the equipment labels.
The control is a capacity change gate. Any addition of load, or any removal of cooling or power capacity from service, passes through a documented assessment signed by facilities and IT, covering thermal (measured inlet map or CFD), power path, redundancy state before and after, and the design ambient assumption. That assessment produces a new declared redundancy state for the hall, which is published and visible, not buried in a ticket. DCIM thresholds should be configured with named actions attached, not just alarms. Time-box every pilot with an explicit exit plan and a load cap, and require accountable-executive risk acceptance whenever an addition consumes redundancy rather than sitting inside it.