← Back to Projects

Scaling to 5k Tenants: Tier-3 Data Center & IPv6 Design

Designed and built two Tier-3 Data Centers from the ground up to support a rapidly growing ISP. Implemented a future-proof Dual-Stack (IPv4/IPv6) BGP architecture.

“Most ISPs were dragging their feet on IPv6 in 2012. We decided to build native dual-stack routing directly into the foundation from Day 1.”

Executive Summary

In 2012, Net4 India experienced massive demand for hosted infrastructure. Our legacy hosting pods were out of power and rack space.

I served as the lead architect for building two Tier-3 Data Centers in Delhi and Chennai from scratch. We designed a high-density Layer 3 Clos Fabric with Dual-Stack BGP (IPv4/IPv6) to support 5,000+ new cloud tenant VMs.

The Challenge

Rapid customer growth exposed capacity limits across legacy flat Layer 2 networks.

  • Rack Space & Thermal Density: Legacy racks suffered from hot spots and unmanaged air recirculation.
  • Spanning Tree Limits: Flat Layer 2 domains created large broadcast storms and limited server density.
  • IPv4 Address Exhaustion: Regional registries were running out of IPv4 allocations, requiring immediate IPv6 readiness.

[!IMPORTANT] Continuing to scale flat Layer 2 networks would have caused catastrophic loops. We needed a scalable, loop-free routing architecture at the top of every rack.

The Solution

We moved away from legacy core-distribution-access topologies to a non-blocking Spine-and-Leaf Architecture.

Technology Stack

  • Edge Routers: Cisco ASR 9006 (Dual RSP440 Control Planes)
  • Core Switches: Cisco Nexus 7010 (NX-OS)
  • Security & Perimeter: Juniper SRX 3600 High-End Firewalls
  • Observability: Observium (SNMP) & RANCID (Config Versioning)

Technical Architecture

  1. Edge Peering Layer: Dual Cisco ASR 9006 routers receiving full internet BGP tables from multiple upstream Tier-1 providers (Tata Communications, Bharti Airtel).
  2. Core Layer: Cisco Nexus 7010 switches running OSPFv3 as the internal gateway protocol for dual-stack routing.
  3. Leaf Layer: Top-of-Rack (ToR) switches handing off redundant 10G uplinks to dual power-feed racks.

[!NOTE] Native Dual-Stack: Every VLAN interface, core link, and customer handoff was dual-stacked with both IPv4 /24 subnets and IPv6 /64 prefixes simultaneously.

Architecture Topology

graph TD
    subgraph "External Transit (Dual BGP Peering)"
    ISP1[Upstream: Tata] --- Edge1[Cisco ASR 9006]
    ISP2[Upstream: Airtel] --- Edge2[Cisco ASR 9006]
    end

    subgraph "Core Spine Layer (OSPFv3 / NX-OS)"
    Edge1 === Core1[Nexus 7010 Core A]
    Edge2 === Core2[Nexus 7010 Core B]
    Core1 === Core2
    end

    subgraph "Leaf & Access Layer (Dual 10G Uplinks)"
    Core1 --- Agg1[Leaf Switch Rack 01]
    Core2 --- Agg2[Leaf Switch Rack 02]
    Agg1 --- Rack[Dual PDU Server Racks]
    Agg2 --- Rack
    end

Business Impact

  • Onboarding Scale: Successfully provisioned infrastructure for 5,000+ VPS and Bare-Metal tenants.
  • Operational Uptime: Delivered 99.999% network availability across the first 12 months of live operations.
  • Compliance Readiness: Exceeded government IPv6 compliance mandates years ahead of competitors.

The Verdict

Key Takeaway

Build Future Capabilities into the Foundation.

Retrofitting dual-stack IPv6 into an existing production network is painful. By embedding IPv6 BGP Peering and Layer 3 Fabrics into the initial physical design, we eliminated future technical debt.