PUE Drift After a Partial Load Migration
A 1.8 MW IT load colocation hall reports annual PUE of 1.42 two years ago and 1.55 last year. The facility team blames hotter summers. Utility data shows total facility energy rose 9% while IT energy rose 2%. The mechanical schedule shows CRAH fans were rebalanced to full speed after a chilled-water valve retrofit, and the economizer was locked out during commissioning and never re-enabled. Sub-metering exists on chillers, CRAH fans, and pumps but not on UPS losses or lighting. The customer wants a single number for the SLA report and a defensible explanation of the trend.
1. How should the analyst decompose the PUE increase to separate weather effects from operational changes, and what data gaps must be closed first?
Read the worked answer
Start by decomposing facility energy into IT energy and overhead, then split overhead by the metered subsystems: chillers, CRAH fans, pumps. With IT up 2% and total up 9%, overhead grew far faster than load, which points to operations rather than weather. A quick check: if PUE went 1.42 to 1.55, overhead per unit IT rose roughly (1.55-1.42)/1.42 ≈ 9% relative, consistent with the fan rebalance plus lost economizer hours. Weather can be isolated by normalizing chiller energy against cooling-degree-days for both years; if CDD-normalized chiller energy is flat, weather is not the driver. The locked-out economizer is the strongest suspect because it removes free cooling. Data gaps: UPS losses and lighting are unmetered, so overhead is understated and the split is incomplete. Before publishing an SLA number, confirm the PUE boundary (does it include UPS and lighting?), fix metering, and re-enable the economizer to establish a clean baseline.
2. The customer insists on a single PUE for the SLA. What caveats and reporting choices should accompany that number?
Read the worked answer
Report the measured annual PUE but state the boundary explicitly (which loads are included), the measurement period, and that UPS and lighting are currently unmetered and therefore excluded or estimated. Note that the value reflects a post-retrofit operating state with the economizer disabled, so it is not representative of design intent or of a corrected configuration. Provide the decomposition alongside the single number so the trend is explainable, and commit to a re-baselined PUE after metering is completed and the economizer is restored. Avoid presenting the number as a like-for-like comparison to the prior year without CDD normalization.