Server room and critical environment monitoring

At a glance

Item	Typical answer
Problem	Server rooms, branch comms rooms and electrical rooms are unmanned; cooling failures, leaks and UPS problems are noticed when equipment shuts down.
What is connected	Rack inlet temperatures, humidity, leak detection, doors, UPS and power distribution values, precision cooling units, generator and transfer switch states.
Who uses it	IT operations, the facilities or data centre engineer, the security desk and the IT manager.
Typical first scope	The main server room plus three to five branch rooms.
Time to first screen	Typically a day for a branch room using the Remote sites solution; one to three weeks for a main room whose UPS and cooling units are read over Modbus.

The situation

A company with a main server room and tens of small comms rooms in branches, factories or stores relies on rooms that nobody sits in. The main room has precision cooling units, a UPS, perhaps a generator; each branch room has a rack, a split air conditioner and a small UPS. Electrical rooms with switchgear and the generator transfer switch sit next door.

The typical incidents: a split unit trips on a Friday night and the branch rack reaches 45 °C by Saturday morning; a condensate pump fails and water spreads under the raised floor; a UPS has been running on a failed battery string for months and drops the load at the first power blip; a door to the electrical room is left unlocked. IT monitoring tools see servers going down, not the room around them, and the building management system, where one exists, does not reach the branches and does not page IT.

The aim: every critical room on one screen, readings close to the equipment's air inlets, alarms that wake the right person for the right reason, and a record that shows what happened before an outage.

Goals and measures

• Inlet compliance = time each rack inlet sensor is within the agreed band ÷ time in the period. A commonly used recommended inlet band is 18 to 27 °C; agree yours with the equipment owners.
• Hottest inlet per room per day from day aggregates, to find hot spots and blanking problems.
• Power usage effectiveness = total facility energy ÷ IT energy, per day, from the increase of the facility meter and the UPS output meter, where both exist.
• UPS runtime margin = minutes of runtime remaining at the current load, and the last self-test result per UPS.
• Cooling redundancy = number of cooling units running and healthy against the number needed for the load.
• Time to acknowledge for Critical alarms, measured from the alarm timeline, in and out of working hours.

Architecture

Server rooms are on corporate networks that rarely allow inbound connections, and their UPSs and cooling units usually sit on a private management network. The design therefore reads everything on site and sends it out over one outbound connection.

• Main room: a licensed gateway running Hermetry Edge on the management network reads the UPS and precision cooling units over Modbus TCP or Modbus RTU through their interface cards, Modbus temperature and humidity sensors on an RS485 line along the racks, and dry contacts on its digital inputs: leak detection controller, door contacts, generator running, transfer switch position and the fire panel's auxiliary fault output. It sends everything over MQTT with TLS on port 8883.
• Branch rooms: a smaller licensed gateway or a licensed 4G router with your own SIM, reading a temperature and humidity sensor, a leak sensor, a door contact and the UPS. Where cabling is hard, LoRaWAN sensors with a LoRaWAN gateway do the same job.
• SNMP: Hermetry's cloud poller can read UPS-MIB, HOST-RESOURCES and interface data over SNMP v2c or v3, but only from a host at a public address with the poller's fixed source addresses allowed. Most UPS cards in server rooms are private, so the gateway path above is the normal one. SNMP polling by the gateway on site is in development.
• Cameras: RTSP cameras and recorders can show live views next to the room data, if the platform can reach them; recording stays in your video system.

Component	Role	Notes
Licensed gateway with Edge	Reads Modbus devices and contacts, publishes telemetry	On the management network; outbound only. Factory identity, claimed into the organisation.
Rack inlet temperature sensors	Air the IT equipment breathes	Front of rack at bottom, middle and top in critical racks; Modbus RTU or LoRaWAN.
Room temperature and humidity sensor	Room air and dew point	One or two per room, away from supply air outlets.
Leak detection	Water under raised floor, near cooling units and pipes	Leak rope with a controller contact, or LoRaWAN spot sensors.
UPS interface card	UPS values over Modbus TCP or RTU	Read-only. SNMP only if the card has a reachable public address.
Precision cooling interface	Unit states, temperatures, alarms	Modbus RTU or TCP card of the unit.
Energy meters	Facility and IT energy	Modbus RTU; needed for power usage effectiveness.
Licensed 4G router	Uplink for branches without a usable line, and a backup path	eSIM management from the platform is in development (Hermetry Connect).

Licensed devices authenticate with X.509 certificates from the platform's own authority; the profile requires TLS and can restrict source networks to your egress addresses. The gateway reports the UPS, each cooling unit, the sensor line and the meters as child devices, so each appears with its own readings, alarm rules and history.

Point list

A main room with two cooling units and one UPS. Register numbers are placeholders; take them from each card's Modbus map. Keys for UPS values follow the names of the platform's SNMP UPS template, so boards look the same whichever path a UPS uses; over Modbus, status values such as the output source arrive as the card's numeric codes, so write rules against those codes.

Point	Telemetry key	Unit	Source	Interval	Notes
Rack inlet temperature, per position	inlet_r01_top_c, inlet_r01_mid_c, …	°C	Modbus RTU sensor, input register 30001	60 s	One key per sensor position; or one child device per rack.
Room temperature	room_temp_c	°C	Room sensor	60 s
Room humidity	room_rh_pct	%	Room sensor	60 s
Room dew point	room_dewpoint_c	°C	Room sensor, if it reports it	60 s
Water leak	leak	bool	Leak controller contact, DI1	on change	Plus a heartbeat every 5 min.
Leak zone	leak_zone	number	Leak controller Modbus, if available	on change	Which section of rope is wet.
Door open	door_open	bool	Door contact, DI2	on change
Fire panel fault or pre-alarm	fire_aux	bool	Fire panel auxiliary output, DI3	on change	Information only.
UPS output source	output.source	text	UPS card	30 s	normal, bypass, battery and others.
UPS battery charge	battery.charge	%	UPS card	60 s
UPS runtime remaining	battery.minutes_remaining	min	UPS card	60 s
UPS battery temperature	battery.temperature	°C	UPS card	5 min	Battery life halves with heat.
UPS output load, phase 1-3	output.load.1, output.load.2, output.load.3	%	UPS card	60 s
UPS input voltage, phase 1-3	input.voltage.1, …	V	UPS card	60 s
UPS alarms present	alarms.present	count	UPS card	60 s
UPS last self-test	test.result	text	UPS card	1 h
Cooling unit running	cu1_run, cu2_run	bool	Cooling unit card	60 s
Cooling unit return air temperature	cu1_return_c	°C	Cooling unit card	60 s
Cooling unit supply air temperature	cu1_supply_c	°C	Cooling unit card	60 s
Cooling unit alarm	cu1_alarm	code	Cooling unit card	60 s	Map codes in the runbook.
Generator running	gen_run	bool	Generator contact, DI4	on change
Transfer switch on generator	ats_on_gen	bool	ATS contact, DI5	on change
Facility energy	facility_kwh	kWh	Main meter	15 min
IT energy	it_kwh	kWh	UPS output meter	15 min

Asset model

• Site per building or branch, with address, location, time zone, site_code, it_contact and facilities_contact.
• Room per server room, comms room or electrical room, with room_type, tier (critical, important, standard), design_kw and inlet_band.
• Machine per UPS, cooling unit and generator, with model, rated_kva or rated_kw, battery_installed and service_contract.
• Other (free kind Rack) for each critical rack, with rack_id and owner_team.

Place each child device on the asset it measures. Add Feeds relations from the UPS to the room and from the transfer switch to the UPS, so a site page can show what depends on what, and Monitors relations from inlet sensors to their racks. Status rolls up: a rack in alarm turns its room and site amber or red on the overview.

Limits that vary by room go on the devices as server attributes, for example inlet_high_c = 27 for a standard room and 25 for a room with older equipment. Rules compare with the attribute; a device that lacks it is not checked for that limit, so include the attributes in commissioning.

Device setup

Start from the Remote sites solution

Branch rooms map well onto the Remote sites solution. It installs a site controller profile with temperature, humidity, door, water leak and smoke keys and sensor batteries, a UPS profile, threshold rules on per-site attributes, offline rules and a UPS watchdog flow that raises a warning when a UPS unit falls silent, plus a board with Sites, Alarms, Site, History, UPS and Camera pages. Adapt its keys to your sensors or map your sensors to its keys. Use separate profiles for the main room: Main room sensor line, UPS (Modbus), Cooling unit, Power and generator.

Provisioning

Claim each licensed gateway by scanning its code and place it on its room. Child devices are created when the gateway first reports them, with the profile taken from the type the gateway sends; place each on its asset. The Recent message online rule with an expected interval of one to two minutes plus an inactivity alarm per profile catches a gateway or child that goes quiet.

Settings and polling

Remote device settings show the gateway model's Modbus addresses, serial parameters and polling intervals. Configure one branch, save a baseline and apply it to all branches with a bulk job in waves; a dry run shows the differences first, and the job stops itself if too few gateways take the change. Poll the UPS output source every 30 seconds so a transfer to battery is seen quickly; inlet temperatures every minute. If the gateway model supports report by exception, use a deadband of about 0.3 °C on temperatures with a heartbeat.

Time and continuity

Synchronise the gateway's clock with your internal time servers; readings more than ten minutes ahead are dropped. When the uplink fails, licensed gateways buffer readings and send them later, within a capacity that depends on the model. Alarm rules evaluate late readings as a backfill: the history shows what happened during the outage, marked Backfilled, and nobody is paged for the past. A branch whose only uplink is the corporate network goes silent exactly when that network fails; a licensed 4G router as a second path avoids that blind spot.

Firmware

Signed gateway firmware rolls out from the profile in waves with soak times; a failed wave pauses the rollout. Devices on 4G wait for metered downloads unless allowed. Begin with the least critical branch.

Dashboards

Board	Audience	Widgets and what they show
Rooms overview	IT operations, security desk	Map of sites coloured by alarm state; table of rooms with hottest inlet, humidity, door, leak, UPS source and last reading; stat cards for open Critical alarms, rooms on battery and devices offline; alarm table. Shown on the operations wall screen through a read-only display link.
Room page	IT and facilities engineers	Inlet temperatures per rack as a heatmap or table; gauges for room temperature and humidity; state cards for door, leak, cooling units and UPS source; line chart of inlets and cooling return and supply over 24 hours; the room drawing; the room's alarms.
UPS	Facilities	Battery charge and runtime as progress cards; output load per phase; input voltages; last self-test result in words; battery temperature trend over 90 days.
Energy and capacity	IT manager	Daily facility and IT kWh and their ratio against the previous period; output load trend against UPS rating; cooling units running.
Cameras	Security desk	Live views of the room cameras next to the door state, if cameras are connected.

Entity parameters keep one room page for all rooms; the overview opens it by drill-down with the room as a parameter. The UPS page takes the UPS device that feeds the chosen room.

SCADA screens

• Room layout: a plan of the room with rows of equipment racks, the precision cooling units and the UPS, from the telecom family; inlet temperatures printed on each rack and coloured by limit; the cooling units showing running, standby and alarm; the UPS showing normal, bypass and battery from output.source.
• Power path: a single-line drawing from the Power pack with busbar, circuit breaker and switchgear cabinet symbols, the standby generator and transfer switch from the Building pack, and the UPS; the active path highlighted by rules on ats_on_gen and output.source.
• Network corner: network switch, firewall and gateway router symbols from the telecom family where a drawing of the comms rack helps the branch technician.
• Reuse: draw a branch room once, attach it as a view of every branch site and let Match everything bind each site's devices. A 3D view made from the room plan keeps its bindings and helps on-site staff find a rack.

See the SCADA library for the packs. Clear the designer's Problems tab before publishing.

Rules and alarms

Alarm	Condition	Delay and hysteresis	Severity	Notified	Escalation
Rack inlet high	any of up to 16 inlet keys > attribute inlet_high_c (Major); > attribute inlet_crit_c (Critical)	Debounce 3 samples; hysteresis 1 °C	Major / Critical	IT operations, push and chat; Critical also SMS to the duty engineer	Rise to Critical notifies again; reminder every 15 min
Room humidity out of band	room_rh_pct > attribute rh_high or < attribute rh_low	Debounce 10 min; hysteresis 3 %	Minor	Facilities, e-mail	Reminder every 60 min
Water leak	leak == true	Debounce 1 sample	Critical	Facilities and IT duty, SMS and push	Reminder every 5 min
UPS on battery	output.source == battery, or the UPS card's numeric code for battery when read over Modbus	Debounce 2 samples	Major	Facilities and IT operations, SMS	Reminder every 10 min
UPS runtime low	battery.minutes_remaining < attribute runtime_min	Debounce 2 samples	Critical	IT duty and IT manager, SMS	Reminder every 5 min
UPS on bypass	output.source == bypass	Debounce 2 samples	Major	Facilities, e-mail and push	Reminder every 30 min
UPS self-test failed	test.result == doneError		Major	Facilities, e-mail	None
Cooling unit alarm	cu1_alarm != 0	Debounce 2 samples	Major	Facilities, push	Reminder every 30 min
Both cooling units stopped	all: cu1_run == false, cu2_run == false	Debounce 60 s	Critical	Facilities and IT duty, SMS	Reminder every 5 min
Door open out of hours	Rules card: door_open becomes true outside the access window		Minor	Security desk, push and in-app	None
Room device silent	Inactivity rule, 5 min		Major	IT operations, push	Reminder every 30 min

Rate of rise. A cooling failure shows as a fast climb long before any threshold. Alarm rules compare values, not slopes; a flow in the room profile's root can compute the change between readings with a delta block and store it as inlet_rise_c with a save block. An alarm on inlet_rise_c > 2 between one-minute readings then warns minutes earlier than the threshold.

Power events without a flood. A mains failure puts every UPS on battery and may start the generator. Keep UPS on battery at Major and let UPS runtime low carry the Critical, so the night page comes only when time is short. The transfer switch and generator states explain the event on the drawing. If more than 20 alarm notifications a minute would go out, Hermetry collects the rest into a five-minute digest.

Lifecycle and records. Alarms open after their debounce and clear themselves. The duty engineer acknowledges and adds notes on the timeline, for example the split unit reset and the ticket number; acknowledgements and clearances are audited under the person's name. Each rule carries a runbook: which breaker, which contractor, which shutdown order. Escalation follows severity: a rise from Major to Critical notifies again, and reminders repeat until someone acknowledges. Hermetry has no on-call rotation for IoT alarms, and alarm rules notify users, roles or e-mail addresses, not groups; route alarms to a shared duty address, or add a Rules-card rule on the alarm that notifies a duty team whose members you change each week.

Maintenance. Mute a room's devices during UPS battery replacement or cooling service. Alarms still record, marked as muted; a mute ends by itself, after at most 90 days.

Automation and integrations

• Ticketing: the signed webhooks alarm.created, alarm.acknowledged and alarm.cleared create and update tickets in the IT service management tool; a flow can instead post only Major and Critical alarms through an HTTP request.
• Chat: alarm notifications can go to Telegram, Slack, Discord or WhatsApp where your organisation has connected them, next to SMS, push, e-mail and in-app.
• Local outputs: a licensed gateway with relay outputs can drive a siren or start an extract fan through a command from its catalog, if the model and firmware support the method; a Rules card rule can send it when the room stays above 35 °C for 5 minutes. High-impact commands need a typed confirmation and are audited. The UPS, cooling units and generator keep their own controls; no commands go to them.
• Remote work on routers: an engineer can open the web panel of a licensed router from its device page through router remote access, in sessions limited in time and audited, without a fixed IP or VPN.
• Reports: history tables download to Excel for monthly capacity and energy reviews. Scheduled or e-mailed reports are not built in today.

Roles and access

Who	Role or group	Can
IT operations	User + IoT operator group	See all rooms, acknowledge alarms, send basic commands, write limit attributes.
Security desk	User	See door, leak and camera views; acknowledge door alarms.
Facilities engineer	User + IoT operator group	See power and cooling devices, acknowledge, add notes.
Branch IT contact	User, narrowed by a policy to their site	See their room only.
IT manager	User + IoT viewer group	Read everything, export history.
Platform administrator	Admin	Profiles, rules, boards, command catalog, mutes, firmware.

Attribute-based policies narrow a member to single devices or the devices of one site. Sign-in can go through your SAML or OIDC identity provider with two-factor authentication or passkeys, which suits an IT department. The AI firewall watches sign-ins, API calls and device traffic, blocks abuse and explains each decision. The audit log records sign-ins, role changes, commands, acknowledgements and setting changes in a hash chain that Verify checks, exportable as CSV or NDJSON.

Sizing and data

As an order of magnitude for one main room and 30 branch rooms:

• Main room: about 8 child devices sending once a minute, and the UPS source every 30 seconds: roughly 8 × 1,440 + 2,880 = 14,400 messages a day.
• Branch rooms: one gateway with two children (sensors and UPS) at one message a minute each, 30 × 2 × 1,440 = 86,400 messages a day.
• Contacts and heartbeats: allow 100 per room a day, 3,100.
• Total: about 104,000 messages a day, roughly 3.1 million a month. Alarms, commands and connection events do not count against the allowance.
• If the branches report every 5 minutes instead of every minute, the total falls to about 35,000 a day; inlet temperatures in a small room rarely need more.

The plan's retention class, from 7 to 1,095 days, sets history length: raw readings are kept for at most 90 days, minute aggregates for up to 31 days, and hourly and daily aggregates for the full class, which make year-on-year capacity charts cheap. The 365-day class, with 90 days of raw data and a year of aggregates, is a common choice for server rooms. Each device's default limits of 10 messages a second and 300 a minute leave plenty of headroom.

Rollout plan

Phase	Weeks	Work	Exit criteria
0. Survey	1-2	List rooms, UPS and cooling models with their interface cards and Modbus maps; network path for outbound MQTT; alarm limits and duty team agreed.	Point list per room type; firewall rule for outbound 8883 approved.
1. Main room	3-5	Gateway, sensor line, UPS and cooling cards; profiles, alarm rules, room drawing, webhooks to ticketing.	Values match the UPS and cooling displays; every alarm forced and handled end to end.
2. Branch pilot	6-7	Remote sites solution adapted; three to five branches; baseline for branch gateways.	Two weeks without unexplained gaps; nuisance alarms under one per room a week.
3. All branches	8-16	Remaining branches in batches; 4G backup paths where needed; display link on the operations wall.	All rooms online; a weekend power-failure drill passed in one branch.

Commissioning checklist

1. Gateway claimed, certificate active, outbound connection on 8883 works through the corporate firewall.
2. Each child device on the right profile and asset; limit attributes present.
3. Each inlet sensor compared with a reference thermometer at the rack front; position labels match keys.
4. Leak rope wetted at its far end: Critical alarm by SMS; clears when dry.
5. UPS switched to battery in a planned test, or its self-test run: output.source changes, the alarm opens, runtime readings fall as expected.
6. Cooling unit stopped in a controlled test: unit alarm, rising inlets, rate-of-rise alarm before the threshold alarm.
7. Door opened out of hours: security desk notified.
8. Uplink unplugged for 15 minutes: offline alarm; buffered readings and backfilled alarms after reconnection; 4G backup path tested where fitted.
9. Ticketing webhook receives created, acknowledged and cleared events with valid signatures.

Pitfalls

• Sensors on the wall. Room air near the door says little about rack inlets. Measure at the front of the racks at several heights.
• SNMP assumed to work from the cloud. UPS cards on private networks cannot be polled from outside; read them through the gateway.
• Every UPS event Critical. Transfers to battery happen during every short dip. Page on runtime, not on the transfer.
• Single uplink through the network being monitored. When the core switch fails, the room goes silent. Add a 4G path for critical rooms.
• No limit attribute on new devices. A device without the attribute is not checked for that limit. Make attributes part of the add-device routine.
• Duty team out of date. Notifications reach people who left. Review the team membership and alarm recipients when the duty roster changes.

What Hermetry does not replace

Hermetry monitors, alerts and records. Fire detection and suppression, emergency power-off, UPS and cooling controls, generator start and transfer logic stay in their own systems and hard-wired circuits and must work without the platform. Server and network monitoring stays in your IT monitoring tools; Hermetry watches the room around them. Video recording stays in your video system.