Zabbix power and temperature monitoring on a 3D room

What to monitor for power and temperature, and what Zabbix shows

Measurement	Where Zabbix usually gets it	What DC Sim does with it
UPS runtime remaining, load %, battery charge %, on battery	a UPS template by SNMP (the standard UPS-MIB or the vendor’s MIB)	runtime, load and charge on the UPS in the one-line diagram
PDU power and current	a PDU template by SNMP: inlet total, per phase, sometimes per outlet	load on the PDU; overloads when the other feed is lost
Server power and inlet temperature	IPMI on the BMC, or the server’s SNMP agent	power on the device; inlet on its rack
Rack inlet temperature and humidity	sensors at the rack front, by SNMP or Modbus	inlet colour on the Air layer
CPU utilisation, ping, interface traffic	the Zabbix agent, ICMP checks, SNMP interfaces	device status and link utilisation on the Network layer

Zabbix dashboards draw these as graphs, top-host lists, maps and geomaps (gauges from 7.0). A Zabbix map is a 2D drawing: a background image, icons placed by hand, links you draw. None of them knows that PDU-R05-B feeds the servers in rack R05, so none can say what happens to those servers when PDU-R05-A fails. That is the gap a twin fills: the same readings, on a model that knows the chain.

1. Build the room first

Zabbix knows hosts and items. It does not know rack positions or which feed powers which PDU (reading rack positions from host inventory is on the roadmap). The readings need somewhere to land, so build the room before you connect:

• From the Excel template: Devices, Racks and Power at least.
• From NetBox: see how to turn NetBox into a 3D data center model.

Use the Zabbix host names as device names, and name the UPS and PDU rows of the Power sheet the way their hosts are named. A UPS host called UPS-A then lands on the UPS-A node of the one-line diagram, and its runtime and load show there.

2. A read-only Zabbix user and API token

• Version: Zabbix 6.0 or later. Below 6.4 the token travels in the request body, from 6.4 as a Bearer header; DC Sim picks the right one from the version.
• User group: Read permission on the host groups that hold your UPSes, PDUs, sensors and servers. Hosts outside those groups are invisible to the token, and so are their problems.
• Role: user type User, API access enabled, API methods limited to an allow list: host.get, item.get, problem.get, trigger.get.
• Credentials: an API token for that user, with an expiry date. A read-only username and password also work.
• Network: a Zabbix inside your network is read by the DC Sim agent, which dials out over one WebSocket and never listens on a port. The token stays in the agent’s configuration; the agent reports open problems every 60 seconds.

3. Which items become readings

The mapping goes by item key and unit, so stock templates and your own items both work without renaming. The key or name only has to tell an inlet from a CPU temperature, and UPS runtime from UPS load.

Zabbix item	Becomes	Shown on
Power in W or kW, on a PDU, UPS or server host	power reading on that node or device	Power layer; tokens per kWh on the Token layer
Current in A	current reading on the PDU	Power layer
Temperature whose key or name says inlet, intake or ambient	inlet reading on the rack	Air layer
CPU or processor temperature	CPU temperature on the device	Device layer
UPS runtime or time remaining	UPS runtime in minutes	UPS node on the diagram
UPS load in %	UPS load	UPS node on the diagram
Battery capacity or charge in %	battery charge	UPS node on the diagram
Humidity in %, fan speed in rpm	humidity, fan speed	Air and Device layers
system.cpu.util, icmpping	CPU utilisation, up or down	Device layer
net.if.in / net.if.out on an interface	link traffic on that port	Network layer
Open problem, severity Disaster or High	critical alarm on the device	all layers
Open problem, severity Average / Warning	major / minor alarm	all layers

DC Sim keeps one value per host and metric, the latest, and a later sync replaces it. Three things follow from that:

1. Set the unit on every item. A power item without W or kW, or a load item without %, is not recognised.
2. Name inlet sensors as inlet. A rack-front sensor whose key and name only say “temperature” becomes a generic temperature, not the rack’s inlet.
3. Give each PDU host a total power item. Per-outlet items on the same host are not added up; the PDU’s load should come from its inlet or total reading. Outlet-level loads come from each device’s own readings or from the catalog’s typical draw.

4. How hosts are matched

A host matches a device or power node when its host name, its visible name or its inventory serial number equals the device’s name or serial in the twin. Preview tells you how many matched before anything is added, for example “4 of 5 hosts match devices · 6 readings · 2 open problems on matching devices”.

Unmatched hosts are listed, and their readings and problems are left out. The usual causes:

• FQDN against short names. pdu-r05-b.dc1.example.com is not PDU-R05-B. Set the visible name in Zabbix, or rename the device.
• Different estates. Our Zabbix test instance (6.0.48, 1,417 hosts) matched none of the demo hall’s devices at first, as it should: its hosts are a different estate. Matching is by name, so the room and the monitoring must describe the same things.
• No inventory. Serial matching needs the host inventory’s serial number filled in, by hand or in automatic inventory mode from an item that populates that field.

5. Connect, Test, Preview, Import

1. Open ⋯ → Connect NetBox or Zabbix, choose Zabbix, paste the URL and the token (or user and password), press Test. Against the test instance, Test reported Zabbix 6.0.48 · 1,417 hosts · 1,375 items · 130 problems in 1.2 seconds.
2. Preview shows the matched hosts, a sample of the readings and the alarms that will be added.
3. Import (“Add readings and alarms”) adds them and creates a new version of the model.
4. Sync again for fresh values: same host and metric are replaced, open problems refreshed. Through the agent, problems arrive every minute and an alarm expires 10 minutes after its last report. On the direct path, alarms are kept for the last 24 hours in memory; keeping them across server restarts is on the roadmap.

No API access? Export recent item values per host and put them in the template’s Readings sheet (time, item, metric, value), with the hosts in the Devices sheet. Three or more times per item and metric are kept as a history.

6. Triggers worth having in Zabbix anyway

The twin does not replace alerting. These four triggers catch most power and temperature trouble before a what-if is needed:

Trigger	Threshold	Why
Rack inlet high	above 27 °C	the top of the ASHRAE recommended range of 18 to 27 °C
UPS on battery	output source is battery	the utility is gone; everything after is runtime
UPS runtime low	below generator start time plus a margin	the survivor must last until the generator carries the load
PDU per feed above 40 % of its rating	on each PDU of an A/B pair	if one feed fails, the other carries the whole rack; above 40 % each, the survivor ends above 80 %

The last one is the 80 % rule applied to failover: a rack on two 16 A, 230 V PDUs (3.68 kW each) must stay under 2.94 kW in total, not per PDU.

7. What a what-if adds on top of monitoring

Zabbix tells you what is true now. A twin answers what would be true after a failure, using what Zabbix measured.

• Fail the UPS. Click UPS-A on the one-line diagram and press Fail. The banner lists the devices that lose power, those left on one feed, the services that stop and every node that overloads, with its percentage. On the demo hall: 2 devices lose power, 90 run on one feed, PDU-R05-B at 113.9 %. With Zabbix readings, those percentages come from your measured watts, not the catalog’s typical draw.
• Runtime on the survivor. Zabbix might show UPS-B at 45 % load with 24 minutes left. After UPS-A fails, UPS-B carries both sides, about 90 %: roughly 12 minutes by the first-order rule (minutes ≈ usable kWh × 60 ÷ load in kW), less in practice because batteries deliver less energy at high discharge. The diagram shows runtime per UPS after the fail, labelled estimated where it is.
• Stop a cooling unit. The scenario gives the minutes until the hottest inlet passes 27, 32 and 35 °C, estimated by a zonal air model, not CFD.
• Replay. Every critical and major alarm of the week sits on the time slider; click one and the room jumps there.

For the full method, from the sheet to the fix list, see the UPS failure impact analysis.