Learn · Explainer · 🏗️
Inside a data center:tiers, power, cooling, racks, cabling — and the jobs that run it.
Before virtualization, before cloud, there is a building with a power chain, cooling, racks and people on shift. Here is the whole stack from the grid to the server PSU, the Uptime tiers, and what each role does.
Tiers: what Tier I–IV actually mean
- Tier I — single path for power and cooling, no redundancy; ~99.67% availability (≈29 hours downtime/year). Server rooms.
- Tier II — single path with redundant components (N+1 UPS, generators); ~99.74%.
- Tier III — concurrently maintainable: multiple paths, only one active; any component can be taken down for maintenance without downtime; ~99.98% (≈1.6 h/yr). Most Indian colocation facilities market Tier III.
- Tier IV — fault tolerant: multiple active paths (2N or 2N+1), compartmentalised; ~99.995%. Banks, exchanges, hyperscalers.
- Tier is a design/certification by Uptime Institute; "Tier III-like" without certification is marketing. Ask for the certificate.
The power chain, from grid to server
- Utility feed(s) (ideally two substations) → HT/LT transformers → main LT panel.
- ATS (automatic transfer switch) decides between grid and DG sets (diesel generators, N+1, typically 24–72 hours of fuel on site, tested weekly, load-bank tested quarterly).
- UPS (online double-conversion; battery or flywheel) bridges the 10–30 seconds until DGs take load, and cleans the power. Sized N+1 or 2N; batteries give 10–15 minutes.
- PDUs (power distribution units) → rack PDUs (A and B feeds) → servers with dual PSUs, each on a different feed. If one whole path dies, nothing reboots.
- Capacity is sold in kW per rack (4–8 kW typical, 15–40 kW for AI/GPU racks) and total IT load in MW. India’s installed capacity grew from 375 MW (2020) to ~1,575 MW (2026).
Cooling
- CRAC/CRAH units push cold air under a raised floor or through ducts; hot-aisle / cold-aisle containment stops mixing and saves 20–30% energy.
- In-row cooling for high-density racks; rear-door heat exchangers and direct liquid cooling for GPU racks above ~30 kW.
- Target: 18–27 °C inlet (ASHRAE), 40–60% humidity. PUE (total power ÷ IT power) is the efficiency score — 1.4–1.8 typical in India, 1.2 excellent.
Racks, cabling and the "rack & stack" job
- Racks are 42U/45U/48U, 600–800 mm wide, 1,000–1,200 mm deep; 1U = 44.45 mm. Servers, switches, PDUs and patch panels are mounted by U position and documented in a DCIM/CMDB (NetBox is the open-source standard).
- Structured cabling: copper (Cat6/6A) for 1–10G, fibre (OM4 multimode, OS2 single-mode) for 10/25/100G and between rooms; patch panels, labelled both ends, colour-coded by purpose. Meet-me rooms hold carrier cross-connects.
- Rack & stack: unbox, rail-mount, cable (power A/B, data, OOB management like iLO/iDRAC), label, asset-tag, record in DCIM, power on, verify with the remote team.
Safety, security and monitoring
- Fire: VESDA early smoke detection + clean-agent suppression (FM-200/Novec/inert gas); no water near IT.
- Physical security: perimeter, biometric mantraps, CCTV, visitor logs, cage/suite access per customer.
- BMS/EPMS monitors power, temperature, humidity, leaks; alarms go to the NOC. ISO 27001 and Tier certifications audit all of this.
The jobs on the floor and behind it
- DC technician / remote hands: racking, cabling, media swaps, reboots, escorting vendors — 24×7 shifts.
- Hardware / field engineer: break-fix on servers, storage, network gear; firmware; vendor RMAs.
- Facilities / M&E engineer: DG, UPS, cooling, fire systems, power capacity.
- NOC engineer: monitoring, first response, ticketing, escalation.
- System / network / storage administrators: the logical layer on top (see role paths).
- DC operations manager: capacity, change control, audits, SLAs, customer onboarding.