Building Two Tier-3 Data Centers from Scratch: PAC Units, Dual Feeders & Fiber Links
How we designed and commissioned two Tier-3 enterprise data centers from empty concrete shells, managing precision air conditioning, dual utility power feeds, and MP-BGP fiber routing.
βAnyone can rack a server. Building a Tier-3 data center means calculating N+1 precision cooling thermal load and Dual-Feeder Automatic Transfer Switches.β
The Setup
In June 2011 at Net4 India, we were tasked with building two greenfield Tier-3 enterprise data centers from empty concrete shells in Delhi NCR. The goal was to deliver 99.982% uptime SLA for hosted enterprise workloads.
Every subsystem β from Precision Air Conditioning (PAC) units to dual utility power feeds, diesel generator synchronization panels, and multi-homed BGP fiber gateways β had to be designed and deployed concurrently.
The Mess
Building physical data center infrastructure leaves zero room for software-style hotfixes. During the initial thermal load test of the primary server hall, 20 high-density server racks triggered an emergency thermal shutdown:
[CRITICAL] 2011-06-14 14:22:05 - PAC-UNIT-02 Alarm
Sensor: Hot-Aisle Air Return Probe #4
Temperature: 38.4Β°C (EXCEEDED THRESHOLD 32.0Β°C)
Status: Compressor #2 Tripped on High Pressure Head Safety Switch
The wrong initial diagnosis blamed defective PAC compressors. In reality, raised floor tile placement created cold-air short-circuiting: cold air returned to the PAC intake without passing through the server chasses, while hot aisles reached 38Β°C.
If we filled the hall with client workloads, thermal runaway would destroy hardware within minutes of a utility power failure.
The Solution
I completely re-engineered the floor airflow dynamics by implementing strict Hot-Aisle / Cold-Aisle Containment and tuning PAC sensor probes:
- Perforated Tile Airflow: Re-balanced raised-floor perforated tiles to deliver 650 CFM per rack directly to cold aisles.
- Containment Blanking Panels: Sealed unused rack units with blanking panels to eliminate hot-air recirculation loops.
- Dual ATS Power Steering: Configured dual-feeder Automatic Transfer Switches (ATS) to fail over between Utility Grid A and Generator Grid B in under 16ms.
# Dual-Feeder ATS & PAC Commissioning Verification Log
ATS-PANEL-01> show status
Utility Feed A: 415V 3-Phase 50Hz (ACTIVE)
Utility Feed B: 415V 3-Phase 50Hz (STANDBY)
Transfer Time Threshold: 16ms
PAC Units 1-4: N+1 Redundant (Air Return Temp: 22.1Β°C - STABLE)
Key Takeaway
Physical infrastructure failure is governed by thermodynamics, not software logic. Always seal hot-aisle containment and validate N+1 cooling thermal dynamics under full load before racking production servers.
Architecture and decisions: mine. Debugging sessions at odd hours: mine. AI assistance: structure, syntax, first draft. β Sachin
Sachin Kumar Sharma
Associate Director (Infrastructure & Cloud Architecture Strategy) | 20+ Yrs Exp
Architecting resilient multi-cloud enterprise landing zones, SDN overlay fabrics, DevSecFinOps automation pipelines, and autonomous Agentic AI platforms.
π‘ Related Engineering Articles
When BGP Local-Pref Steered Us into a Routing Loop
An unredacted post-mortem of how a routine multi-homed BGP engineering change window triggered a transit routing loop and took down a Tier-3 datacenter.
The Edge Intelligence Curve: 135M vs 500M Browser LLMs
Analyzing latency, VRAM allocation, and domain accuracy differences between SmolLM2 and Qwen2.5 running locally on WebGPU.
Token Economics: Applying BGP AS-Path Prepending Principles to LLM Model Steering
How we adapted 15-year-old BGP WAN routing principles (Local-Pref, AS-Path Prepending, MED) to route multi-agent LLM prompts dynamically across cost and latency tiers.
π¬ Stay Updated on Tech Releases
Sign up to get notified when I publish new production war stories, agentic AI architecture blueprints, or open-source infrastructure tools.