Network Architecture

Data Center Network Architecture Guide

Complete guide to designing robust data center networks - Spine-leaf architecture, redundancy, bandwidth planning, security, and best practices for Indian data centers.

Network Architecture Design

Network Topologies

Three-Tier Architecture

Traditional hierarchical design with core, distribution, and access layers.

  • Core → Distribution → Access
  • Simple to design and manage
  • Limited scalability
  • Higher latency between tiers

Spine-Leaf Architecture

Modern, scalable design with full mesh connectivity between spine and leaf switches.

  • Full mesh connectivity
  • Predictable latency
  • High scalability
  • Recommended for modern DC

Fabric Architecture

Unified fabric with single management plane for compute, storage, and network.

  • Single management plane
  • Automated provisioning
  • Reduced operational overhead
  • Higher initial cost

Switching Requirements

Switch Type Port Speed Port Density Throughput Recommended Brands
Core Switches 100/400 GbE 32-64 ports 12.8-25.6 Tbps Cisco Nexus 9K, Arista 7280
Spine Switches 40/100 GbE 36-48 ports 3.6-7.2 Tbps Cisco Nexus 3K, Juniper QFX
Leaf Switches 10/25/100 GbE 48 ports + 6 uplinks 1.2-2.4 Tbps Cisco Nexus 2K, Dell S5200
Top-of-Rack 10/25 GbE 48 ports 480 Gbps-1 Tbps HPE Aruba, MikroTik

Bandwidth Planning

Application Type Bandwidth per Server Redundancy Factor Total for 100 Servers Recommended Design
Web Hosting 100 Mbps - 1 Gbps 2:1 oversubscription 5-50 Gbps 10GbE ToR, 40GbE uplinks
Database 10-40 Gbps No oversubscription 1-4 Tbps 25/100GbE leaf-spine
Virtualization 25-100 Gbps 4:1 oversubscription 0.625-2.5 Tbps 100GbE fabric
Storage 10-100 Gbps No oversubscription 1-10 Tbps Separate storage network

Redundancy Design

Network Level Redundancy

  • Dual ISPs with diverse paths
  • BGP routing for failover
  • Multiple edge routers
  • Geo-redundant connections

Device Redundancy

  • Pair of core switches (VRRP/HSRP)
  • Dual firewalls (active-active)
  • Multiple load balancers
  • Redundant power supplies

Path Redundancy

  • Multiple fiber routes
  • Diverse entry points
  • Mesh connectivity
  • ECMP (Equal Cost Multi-Path)