Power Draw Per Circuit
Back to Data Center Management · Sensor Coverage · Threshold Alerting · Historical Trending · Service Offerings
This is tracking PDU load so no circuit is quietly running near its limit. The word doing the work is quietly. A circuit approaching its rating gives no warning of any kind — no noise, no heat you would notice, no degradation. It behaves perfectly until the breaker opens, and then an entire rack is dark at once.
1. You Have Less Headroom Than the Label Says
A breaker's rating is not a budget you may spend. Electrical codes treat a load running for three hours or more as continuous, and require the circuit to be sized at 125 per cent of it — which in practice means limiting continuous draw to 80 per cent of the rating. Everything in a data centre is a continuous load.
| Breaker | Usable continuous | At 230 V, single phase |
|---|---|---|
| 16 A | 12.8 A | ≈ 2.9 kW |
| 20 A | 16.0 A | ≈ 3.7 kW |
| 32 A | 25.6 A | ≈ 5.9 kW |
Plan against the middle column. A rack designed to 16 A on a 16 A breaker is a rack designed to trip.
2. Measure Per Phase, Not Per PDU
On a three-phase PDU the total is an average across three independent circuits, and an average hides the one that matters. The illustration above is the common case: the PDU reports 67 per cent and looks healthy, while phase L2 sits at 92 per cent and is the thing that will actually trip.
Imbalance is normal and must be managed. Equipment is added one unit at a time, rarely in threes, so phases drift apart unless someone is watching. Rebalancing means moving cords between outlets on different phases — cheap to do, impossible to plan without per-phase measurement, and much easier when the outlet-level cabling is documented.
3. The Redundancy Arithmetic Nobody Does
This is the error with the largest consequences. A rack on A and B feeds is supposed to survive losing one. For that to be true, either feed alone must be able to carry the whole load.
So the steady-state ceiling per feed is not 80 per cent. It is half of that: roughly 40 per cent of the breaker rating on each side, so that when one fails the survivor lands at 80 per cent rather than 160.
Filling both feeds to 80 per cent is extremely common, because each one individually looks correct on every dashboard. The failure then cascades: feed A drops, feed B instantly doubles, exceeds its rating, and trips — and the rack that was built for redundancy goes dark because it was dual-fed. The corresponding physical check, that dual-PSU equipment is actually on two different feeds, belongs with the rear rack elevation.
4. Inrush, and Why Recovery Trips Breakers
Steady-state draw is not the peak. When power is restored after an outage, every power supply, every disk spinning up and every fan at full speed starts simultaneously, and the combined inrush can be several times the running load. Racks that ran for years without incident trip on the way back up.
- Stagger restoration, by PDU outlet group or by delayed start in the BIOS.
- Know the measured peak, not only the running average — which requires a sampling interval fine enough to see it.
- Rehearse it. A power recovery that has never been tested is an assumption, and this belongs in disaster recovery planning.
5. Nameplate Is Not a Capacity Plan
Power supply nameplate ratings describe the maximum the supply can deliver, not what the machine draws. Real draw is routinely 30 to 50 per cent of nameplate and varies with load. Planning from nameplate leaves a great deal of capacity unused; planning from measurement is the whole point of metering.
Two further details worth getting right:
- kW and kVA are not the same. Breakers and UPS capacity are limited by current, which follows kVA; electricity bills and heat follow kW. With modern supplies the power factor is close to one and the gap is small, but it is not zero, and a UPS sized in kVA against a load measured in kW is a sizing error.
- Heat follows power. Essentially all of it. Every watt drawn is a watt the cooling has to remove, which is why this page and sensor coverage are two views of one problem.
6. What to Record
- Per-circuit and per-phase current, with the breaker rating and the 80 per cent figure alongside, so a reading is interpretable without arithmetic.
- Peak as well as average, at a sampling interval that can see inrush.
- Per-outlet draw where the PDU supports it — this is what lets consumption be attributed to an asset rather than a rack.
- Which feed each cord is on, so the redundancy check in section 3 can be run at all.
- Rack and room totals against the room's provisioned capacity, feeding capacity trending.
How We Approach It
- Establish what is actually drawn, per phase and per outlet where the hardware allows, rather than working from nameplate figures.
- Set the real ceilings — 80 per cent of rating for single-fed, and the halved figure for anything dual-fed that is meant to survive a feed loss.
- Run the redundancy check across every rack and report which ones would trip on losing a feed. This finding alone usually justifies the exercise.
- Report phase imbalance and produce a rebalancing plan of cord moves.
- Capture peak and inrush behaviour, and define a staggered restoration order.
- Hand the thresholds to alerting and the series to trending, so that creeping draw is caught months before it matters.
What You Get
- Per-circuit and per-phase draw, measured, with each reading set against its real usable limit.
- A list of every rack that would trip on losing one feed, with the headroom figure for each.
- A phase rebalancing plan expressed as specific cord moves.
- Measured peak and inrush, and a staggered power-on order that has been tested.
- Alert thresholds and trend series wired into your existing monitoring.
The question this work answers is the one that is embarrassing to be unable to answer: can this rack take one more server, and what happens to it if a feed drops while that server is running?