Blog

EVPN multihoming vs MLAG: what actually breaks Both make two leaf switches look like one LACP partner to a dual-homed server. Both forward on both links. So which one holds up when something fails? Two labs on Cumulus Linux, identical wiring, four failure scenarios, every probe accounted for.

A MAC address leaves a Cisco datacenter and reappears behind a Juniper leaf in another one. Watch RFC 7432 do its job across the multisite DCI: the counter climbing 0, 1, 2, the losing leaf yielding, and the exact BGP packets that carried each move, captured on the interconnect.

Type-2 gets all the attention, but a datacenter fabric runs on five core EVPN route types and quietly depends on every one. Here is what each actually carries, decoded from a live multihomed fabric, including the two types MLAG labs never get to show you.

One git clone, four commands, and 20 minutes later a client curls an anycast VIP by DNS name through two firewalls into whichever datacenter the control plane picked. 39 nodes, zero manual configuration. Here is how the whole lab became a git repo.

Part 7 of the Agentic NOC series. How you earn the right to trust a system that watches and sometimes changes your network: 131 offline tests against real captured output, mutated-data fault detection, a live 30-fault campaign, and the worst bug of all, a monitor that mistakes silence for success.

Part 6 of the Agentic NOC series. Real captured runs: a request crossing both protocol layers to a live switch, the planner fanning out, a 30-fault detect-remediate-verify loop, the monitor self-healing then escalating on a flap and refusing to fake health.

Part 5 of the Agentic NOC series. Three read-only MCP servers over a live fabric, firewall pair, and cluster; three A2A specialists and a supervisor; a local-first planner; one guarded write path; an autonomous monitor. The one design law, and how to rebuild it.

vJunos on containerlab is routine on Intel. On an AMD EPYC host the stock image refuses to start its nested VM, and the first test topology most people write hides a second failure that looks like AMD and is not. What breaks, what current vrnetlab already fixes, all reproduced on an EPYC 9575F.

Put the vJunos 26.2 images on an AMD EPYC box running EVE-NG: the switch and router power themselves off, patch that and they hang at mount root, and Evolved on the default AMD template never starts QEMU. Four small edits in two places, each reproduced on the EPYC, every command ready to copy.