Design and Implementation of an Automated Spine-and-Leaf Data Center using Infrastructure as Code

Abstract

   Modern data centers require fast and reliable networks to handle massive amounts of information; thus, traditional three-tier designs are rendered inefficient for heavy east-to-west traffic. To address this, the Spine-and-Leaf architecture has been adopted to provide faster, and highly predictable connectivity. Because the manual configuration of these network switches is slow and prone to human error, an Infrastructure as Code (IaC) approach was utilized to completely automate a modern data center network. A scalable Spine-and-Leaf network was designed in a virtual lab using Arista vEOS switches. A physical underlay was established using OSPF and IS-IS, and a virtual overlay network was created utilizing MP-BGP EVPN and VXLAN technologies.

The entire network was provisioned using automation tools, specifically Ansible playbooks and Python scripts. Following successful deployment, key network metrics, including latency, throughput, and jitter, were measured. A comparative analysis was performed between a baseline 2-leaf topology and a scaled 4-leaf topology, proving the automated network can scale smoothly to handle increased traffic without losing performance. Finally, a self-healing automation system was successfully developed. A custom dashboard was engineered to continuously monitor network health in real-time, executing remediation scripts to fix failures within sixty seconds without human intervention.