Skip to content
Register

Documenting your actual power draw

Datacenter Talk by celiadog 11 replies 2.1K views
celiadog

Member OP

Edge Obsessed

2 #1

I have pop in many facilities and the origin of most outages is: nobody knows their real draw. Spec sheets lie. Your "idle" is my miss.

Setup: cheap smart PDU from any vendor, snmp to a tiny box, log to sqlite. I run this at every pop. It's 12 now (latency to insight, not ms, the number of minutes to catch a thermal runaway).

Scripts below for pulling kW and pushing to whatever dashboard you origin. I will cache hit this with Grafana templates if people want.

What numbers are you actually seeing versus spec?

PU

Member

#2

Will this break my setup if I plug the smart pdu in wrong?? I have a small rack at InterServer and I dont want to knock anything offline

Also where does the tiny box go, just any shelf?

pieter_rtm

Member

#3
  • Raspberry Pi 4: 7.5W peak, $35
  • Script: 200 lines Python, pulls SNMP every 30s
  • SQLite: zero config, survives unclean shutdowns

My machines at ovhcloud.com run 340W spec, 410W measured. As said, fans and RAID controllers hide in idle.

Grafana template attached. Dry fact: most "idle" gear is 40% of peak, not 10%.

Containers before it was cool
lucdump

Member

#4

The smart PDU is good idea but I have try with the Hostinger unit and the reading is all wrong. Ost says 2.3A, my clamp meter says 3.1A. This is not possible to do??

I have make script anyway, it pull from the PDU and the clamp both, now I have two lies instead of one. Bisous

Lucdump

herebungkus

Member

🌏 SEA represent

#5

40% idle sounds high. My gear at home maybe 25%?

wrap it, ship it, pray it
lookuppierre

Member

#6

Hostinger clamp off too? I had this, the PDU was fine actually

prix fixe infrastructure: €5/mo
PetraSuper

Member

network boot believer

#7

Which snmp version, v2c or v3?

boot anything, anywhere, anytime
celiadog

Member OP

Edge Obsessed

#8

PetraSuper is right on v2c. I have thrown away more "smart" PDUs for v3 bugs than for bad calibration. The SNMP stack on these things is written by someone who read a tutorial once.

lucdump said:
Ost says 2.3A, my clamp meter says 3.1A

Clamp is probably closer but not gospel. Power factor on switching PSUs is 0.6-0.9, clamp meters assume 1.0 unless true RMS and even then. You need both voltage and current sampled together, which your PDU might not do. The 2.3A vs 3.1A gap is real but the truth is neither number.

I log both, diff them, alert when diff grows. Trend is truth, point is lie.

lucdump

Member

#9

Thank you celiadog, I have learn something. My clamp is not true RMS, it is 15 euro from Amazon. The Hostinger PDU is maybe not so bad then, or both are bad in different direction.

I will look for true RMS clamp or just trust the trend as you say. Bisous

Lucdump

pieter_rtm

Member

#10
herebungkus said:
40% idle sounds high. My gear at home maybe 25%?

Home gear is different. Single PSU, no BMC, no backplane expanders, no RAID cache battery trickle. My OVHcloud machines have dual PSU and both draw even if one carries load. Your home NAS is not comparable to a server chassis.

I measured my home router + NAS: 23% of peak. But that is not what this thread is about.

Containers before it was cool

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft