What do you use for environmental monitoring that doesn't need cloud?

I frequently encounter this requirement in regulated deployments. Below is my current recommendation matrix for air-gapped environmental monitoring.

• Temperature/Humidity: APC NetBotz 750 with local SNMP trap forwarding
• Leak detection: RLE SeaHawk with dry contact to NetBotz
• Smoke: Vesda VLP with relay output
• Door contact: Generic dry contact, NetBotz input
• Central collection: LibreNMS or Nagios Core on Debian, no outbound connectivity required

Critical constraint: several sensors with Chinese-manufactured components face ITAR-adjacent scrutiny for US government-adjacent workloads. I advise verifying component country of origin in procurement documentation.

Someoneelses compliance department run amok again. The NetBotz 750 has a Mainboard manufactured in someoneelses Shenzhen facility and suddenly that's a National Security Event because your customer has a .gov email address.

The whole ITAR thread here is going to be hilarious because nobody reads the Commerce Control List. Environmental sensors are EAR99, not ITAR. You're all panicking about the wrong Export Administration Regulations.[1]

But sure, let's all pretend that a Temperature Probe is a Munition. Capitalized Bureaucratic Theater at its finest.

[1] I actually called BIS last tuesday. Confirmed. Stop spreading FUD.

(look, I love a good compliance spiral (who doesn't (especially at 2am before an audit))) (but @Berlin1 has a point here (maybe (I'm as surprised as you are))) (the real issue isn't ITAR (it's that your "chinese government customer" probably means PRC state-owned enterprise (which triggers completely different restrictions (CFIUS-adjacent (not ITAR))))) (so everyone's yelling about the wrong thing (classic)) (in my experience (which is, admittedly, mostly watching other people panic (then cleaning up their ansible playbooks (because of course they used shell modules for everything (don't get me started)))))

The SNMP part though — LibreNMS polling every 30 seconds with default timeouts? (you'll miss thermal events (ask me how I know (datacenter baked (literally (45C intake (we "saved money" on monitoring)))))

APC NetBotz 750 is Shenzhen-made lol

This. Exactly this. I was sloppy with my shorthand. The actual problem my clients hit is CFIUS/EAR entity-list stuff when the end customer is SOE-adjacent, not ITAR proper. Thanks for the correction.

Still, the procurement documentation dance is real. I have a customer in The Hague who had to rip and replace a full NetBotz deployment because the integrator couldn't produce component-level COO paperwork. Not because the sensors were illegal. Because the compliance officer couldn't tick the box.

Rip and replace because of paperwork lol

Of course you did. I once had to explain to our procurement that "designed in USA" and "manufactured in USA" are different fields on the form. They put same answer in both. Audit was fun.

For air-gapped monitoring I use PRTG on-premise with Modbus TCP sensors from Comet System. Czech company, EU origin, no cloud dependency. Not as pretty as NetBotz but paperwork is trivial.

What is wrong with 30 seconds? I run LibreNMS for my two racks in Lyon, polling every 60 seconds, zero problems. The default timeout is 1 second yes? For SNMP over local network this is plenty. You want faster, you pay for something else.

Self-hosted Zabbix is also fine. I use both. Zabbix has better mapping for datacenter floor plan.

Nothing (if you like stale data (during a thermal runaway (in a 42U rack (where the difference between "fine" and "melting" is about 90 seconds (ask me how I know (no actually don't (the NDA has teeth)))))))

I run 10-second intervals with 500ms timeout (custom poller (not LibreNMS core (wrote it in Go (because of course I did)))). LibreNMS default is fine for "is the building still standing" monitoring. Not for "did the CRAC just fail" monitoring.

Also ( @harbourops ) — your matrix misses vibration. We had a generator test schedule drift (someone set it to weekly not monthly (someone else (don't look at me (okay look at me)))) and the rack resonance killed two disk arrays in six months. NetBotz doesn't do seismic. We added Wilcoxon 786As with a Raspberry Pi and custom alerting.

Fair point. I typically spec that separately under mechanical infrastructure, not environmental, but you're right that it belongs in a complete picture. The Wilcoxon + Pi approach is clever for cost, though I'd worry about long-term supportability if the person who wrote the poller leaves.

I've started looking at AKCP sensorProbe+ for sites that need the full stack in one box. Singapore company, but they publish full COO chain and have EU assembly option for government work. More expensive than NetBotz, significantly.