1. Why we built this
We wanted a straight answer to a question that comes up every time someone proposes SRv6: can you take a live L3VPN that runs across Nokia, Cisco and Juniper on an SR-MPLS core, and move it to SRv6 one router at a time without dropping customer traffic. The vendor slideware all says yes. We wanted to see it on real control planes, with a customer ping running the whole time, and write down exactly what happened.
This document records what we built, every problem we hit, and the evidence for each claim. It is deliberately heavy on the failures, because in a multivendor build the failures are where the useful information lives. The headline is honest and more interesting than a clean “it works”: the SR-MPLS L3VPN interoperates across all three vendors and runs fine, the make-before-break primitive that a hitless migration depends on works perfectly on the Nokia PE, and it does not work on the Cisco and Juniper PEs on the exact software we tested, for two different and very specific reasons. Both of those reasons are things a vendor needs to hear.
2. Topology
Ten routers in a flat, meshed core. Three of them are provider edge, one per vendor, plus a route reflector. Each PE has a customer site behind it, and all three sites sit in the same VPN, CUST-A.