- Building management system (BMS)
- The real-time control layer — the building's reflexes — that reads sensors and drives actuators (pumps, fans, dampers, breakers) to hold power, cooling, and environment within safe limits second by second. Conservative, industrial software supplied by established automation vendors.
- DCIM (Data Center Infrastructure Management)
- The awareness and planning layer that aggregates and models information about the whole facility — racks, power provisioned versus drawn, cooling and space available — to support capacity decisions over hours to years, prevent stranded capacity, and translate the facility's state into economic terms.
- Facility layer (nervous system)
- The physical infrastructure and control software that runs a data center's power, cooling, and monitoring — the 'second machine' that keeps the building alive, distinct from the compute workload it houses. Governed by operational technology and treated in this course as its own intelligence surface.
- Liquid cooling (direct-to-chip and immersion)
- Cooling methods that bring liquid close to the heat source because liquid carries heat far more effectively than air. Direct-to-chip pipes coolant through cold plates on the hottest components; immersion submerges whole servers in non-conductive fluid. Adopted for high-density compute that air can no longer cool, requiring new facility infrastructure.
- Operational technology (OT)
- Control systems that sense and act on physical processes — power, cooling, environment — in real time, descended from industrial and building automation. Distinct from information technology (IT), which runs the compute workload. OT prioritizes determinism and safety over features and throughput.
- Power Usage Effectiveness (PUE)
- A facility efficiency metric: the ratio of total energy drawn by the building to the energy delivered to the compute. A value near 1.1 indicates little overhead; higher values indicate more energy spent on the facility itself, mostly cooling. It measures the building, not the servers.
- Redundancy (N, N+1, 2N)
- The practice of building spare capacity and parallel paths into power and cooling so no single component's failure stops the flow. N is exactly the required capacity; N+1 adds one spare unit; 2N duplicates the entire path. Greater redundancy absorbs more failure at greater cost.
- Stranded capacity
- Capacity that physically exists but cannot be sold because a complementary resource is exhausted — for example, power available in a zone whose local cooling is full, or vice versa. Because usable capacity requires power, cooling, and space together, imbalance turns paid-for capacity into lost revenue.