A 48-rack enterprise data center built on a non-blocking spine-leaf fabric — 12 spine switches, 48 leaf switches, dual-fabric SAN storage networking, and the worked calculation showing exactly why this fabric can't reuse the campus-style hierarchical design from Project 1.
The project brief: 24 compute racks and 4 storage racks, roughly 800 physical servers, networked on a strictly non-blocking (1:1) Clos fabric rather than the oversubscribed three-tier hierarchy a campus building can safely use. Data center traffic is dominated by east-west flows — server-to-server, host-to-storage, VM migration — that a hierarchical core would bottleneck, so this project builds a spine-leaf topology where every leaf-to-leaf conversation is a constant two hops, sized by a worked calculation that derives exactly 12 spine switches from the leaf configuration's downlink capacity.
This module walks the whole project end to end: spine-leaf (Clos) fabric fundamentals and ECMP/BGP underlay design, the full worked non-blocking sizing calculation, the two independent levels of redundancy (dual-homed servers at the leaf, ECMP across spines at the fabric), the dual-fabric Fibre Channel SAN serving the storage racks, and exactly what distinguishes this project's traffic-pattern assumptions from Project 1's. The full worked numbers and complete bill of materials are part of the unlocked module below.