Examples · browser simulation
The servers have power. Where does the heat go?
Explore a fictional data-center room where compute, cooling, water flow, and backup power have different dependencies. Watch a small failure consume thermal headroom even while compute remains powered.
All values and thresholds are fictional. No connection to a data center or the private lab. One playback second represents one model minute; refreshing discards the session.
Choose an experiment
Two 60 kW cooling trains serve an 80 kW example workload. Start the clock, change demand, or inject a fault.
Each experiment button resets the room, time, faults, and UPS reserve. Fault controls below let you combine failures within a run.
Elapsed 0 model minutes · Paused. Advance pauses playback. Pause stops physical evolution and UPS use in this model.
24.0 °C
Within example limits · warning 28 / critical 34 / stop 38 °C. These are teaching thresholds, not equipment specifications.
80 / 120 kW
Current generated heat / maximum available cooling. Below capacity, cooling modulates toward 24 °C. Capacity is not the same as actual heat removed.
Running
Compute has sustained supply.
This teaching display remains accessible independently of the fictional room.
Follow the cooling dependency chain
Facility supply → heat rejection and shared water flow → two cooling trains → heat removed from the room. The two train cards are parallel branches, not one train feeding the other.
- Facility power✓ Utility available
- Heat rejection plant✓ 120 kW assumed available
- Shared water loop✓ Flow available
- Cooling train A✓ 60 kW available
- Cooling train B✓ 60 kW available
Compute has a separate, IT-only UPS: 8 of 8 model minutes remaining. Runtime is fixed regardless of demand in this simplified model; recharge and standby loads are not modeled.
Thermal response takes time
Temperature changes from the previous interval’s heat imbalance. Restoring a dependency changes capacity immediately; cooling the room takes subsequent model minutes.
Change demand or inject a fault
Changing demand does not restart a stopped workload. At 30 kW a single 60 kW train can cover it; at 80 kW it cannot.
Recover the dependencies, then the service
- ✓ ReadySustained facility power
- ✓ ReadyWater flow and enough cooling for selected demand
- ✓ ReadyExample temperature at or below 27 °C
Recovery check: not accepted for restart. A configuration or demand change invalidates it. Real recovery also needs human, equipment, and site-specific checks absent here.
Event notebook
Two 60 kW cooling trains serve an 80 kW example workload. Start the clock, change demand, or inject a fault.
- 0 min Two 60 kW cooling trains serve an 80 kW example workload. Start the clock, change demand, or inject a fault.
Model assumptions and the heat-balance equation
Each minute: temperature change = (IT heat − heat removed) ÷ 20, with a fictional thermal capacitance of 20 kW·min/°C. Cooling modulates to a 24 °C floor. Each train has 60 kW maximum capacity and shares one water pump and a 120 kW heat-rejection plant. Plant availability is assumed whenever facility power exists. Facility backup is assumed adequately sized and takes two model minutes to start. The eight-minute UPS supports compute only.
The model treats the room as one temperature, omits outdoor conditions, airflow distribution, humidity, pump curves, standby/residual heat, water use, battery-load effects, fuel limits, and real transfer/control logic. Losses can be due to many causes; a failed component here is not an attribution to a cyberattack. The 28/34/38 °C thresholds and 27 °C restart criterion are arbitrary teaching policies, not ASHRAE limits or a real protection design. No real resilience rating or N+1 compliance is established.
CSV temperature and component state describe the end of the step; heat removed describes the interval just completed. History retains 180 samples and the trend shows 60 model minutes. Nothing is persisted. Playback pauses when the tab is hidden, and paused model time does not consume UPS reserve.