Network engineering guides

Configure a 2-spine / 4-leaf CLOS fabric from scratch via the in-browser console. Learn shared-AS spines, per-leaf ASNs, ECMP with multipath-relax, peer-groups, fast BGP timers — and why each choice matters for AI workloads.

Stretch a single L2 segment (VNI 10100, subnet 192.168.100.0/24) across all four leaves on top of the working underlay. Bring up Linux bridge + VXLAN devices, activate the BGP L2VPN-EVPN address family, watch Type-2 MAC routes traverse the fabric, and confirm overlay traffic actually flows leaf-to-leaf.

Bring the eight GPU workers onto the stretched L2 segment you built in Lab 2. Attach each leaf's worker ports as VLAN 1000 access ports, move every worker's eth1 onto 192.168.100.0/24, and run a real Gloo AllReduce — first 2-rank across two leaves, then full 8-rank across the fabric.

Bring up a gNMI streaming-telemetry pipeline (gnmic → Prometheus → Grafana) over your working fabric. Watch real-time per-link Mbps fill in as you run AllReduces — see ECMP load-spread across both spines, see worker ports light up, build operator intuition for what a healthy AI training fabric looks like in motion.

Walk the math behind multi-pod AI fabrics: when does a 2-tier CLOS run out of room, what does the third tier look like in BGP, and why do hyperscalers schedule training jobs to live inside a pod when they can? Conceptual lab — no super spines are deployed; you inspect your existing fabric to anchor each point.

Stand up the IPv6 dual-stack underlay that the SRv6 transport will ride on — additive alongside the IPv4 underlay and EVPN-VXLAN overlay you already built. With the /127 link addresses pre-set, configure a second numbered IPv6 BGP session on every switch — the spines and the leaves — advertise each leaf's uSID locator /48, and confirm every locator is reachable via both spines with per-flow ECMP. No SRv6 dataplane yet — that's the next lab.

Starting from the unnumbered IPv6 underlay + ECMP you built in Lab 5, lay a Segment Routing over IPv6 (SRv6) uSID transport across the fabric. Define micro-SID locators in FRR, program kernel End.DT6 endpoints so each leaf decapsulates, steer flows into uSID with an H.Encaps headend — then watch distinct flows spread per-flow across both spines on the IPv6 flow label, the SRv6 analog of the VXLAN ECMP you saw in Lab 4.

Stand up a compact enterprise EVPN-VXLAN pod — 2 spines, 2 leaves, 2 PCs — and watch two hosts on different leaves reach each other across a stretched L2 overlay. Same EVPN-VXLAN building blocks as the AI fabric, at campus scale, on its own dedicated topology.

Carve the campus access layer into per-device-class VLANs — workstations, IP phones, PoE cameras, and wireless APs — each stretched as its own EVPN-VXLAN segment across both leaves. Same-VLAN devices reach each other across the fabric; the classes stay isolated from one another (no inter-VLAN routing yet — that's the next module's anycast gateway).

Give the segmented campus a distributed anycast gateway — a per-VLAN SVI carrying the same gateway IP and MAC on every leaf — so each endpoint's default gateway is local on its own switch and the device classes can route to one another, same-leaf and across leaves, over the EVPN-VXLAN fabric.

Hand out addresses dynamically: turn each leaf into a DHCP relay agent that forwards the access endpoints' requests across the L3 boundary to a central DHCP server, and watch workstations, phones, cameras, and APs lease their address — and the right gateway — through the relay.