The Watts Hidden in Volt-Amps [/ðə ˈwɑts ˈhɪdən ɪn ˈvolt-amps/] n - The symptom. A row of Rack power distribution units reported branch loads that, summed, exceeded what the upstream Power distribution unit metering said the whole row was drawing. Because capacity planning trusts the per-rack numbers, several racks were flagged “nearly full” and held back from new servers; we were stranding real, paid-for capacity on paper.

The dead end. [/ðə ˈdɛd ˈɛnd/] n - The easy story ran “a sensor is bad, replace the PDU.” Yet swapping a unit demands a maintenance window and risk to a live load, and the numbers were consistently high, not randomly noisy; a failing sensor drifts or spikes, it does not lie by a steady amount.

The actual cause. [/ðə ˈæktʃuəl ˈkɔːz/] n - The rack PDUs reported apparent power (volt-amps) per branch, while the upstream PDU reported real power (watts). On loads with a power factor below one, apparent power is ever the larger number. We were comparing VA to W and treating the difference as load. The metering was not broken; we were adding two different quantities.

The fix - limited and clever. [/ðə ˈfɪks/] n - No hardware. I pointed my Power-Guardian poller at both layers over SNMP, pulled the power-factor field the rack PDUs were already exposing, and normalized every branch reading to real power before it reached the dashboard. The “overloaded” racks at once showed their true headroom, and the stranded capacity returned without anyone touching a breaker.

The lesson. [/ðə ˈlɛsən/] n - Half of capacity work is making sure two numbers mean the same thing before you subtract them. This is also why the Crafts tooling now logs the units alongside the values.

Part of Field Notes.