Summary

  • At RIPE 92, Benedikt Stockebrand suggested moving some network functions from Layer 2 to a routed IPv6 underlay, including VXLAN in place of troublesome VLAN arrangements.
  • His sharpest target was spanning tree: one extra hop could leave a customer's large network failing strangely, without useful error messages. The public talk offers a design case, not measured results from that customer.

Spanning tree has an excellent publicist: it lets a network keep pretending that more cable and more switches are still one manageable room. Benedikt Stockebrand described the moment that story ended. At a customer's network, adding one hop to an already large topology made failures erratic. The operator was left hunting a ghost, he said, because the protocol offered little help explaining what had broken.

He made the case at the RIPE 92 IPv6 Working Group on 21 May 2026. The talk's playful title promised a “killer app” for IPv6. Its actual target was the stack of Layer 2 devices and rules that operators keep maintaining while waiting for an address-family conversion to justify itself. Stockebrand's answer was to route earlier and ask which old mechanisms could leave.

That is a specific engineering claim. In the non-core switching context he described, modern Layer 3 switches could route as quickly as they switch. That removes a traditional performance excuse for preserving some Layer 2 designs. He did not publish benchmark data, switch models or a customer topology. The claim deserves a lab comparison, especially before a purchase order turns a talk into policy.

VLANs were one target. Stockebrand suggested VXLAN over IPv6 when a VLAN arrangement starts imposing limits. RFC 7348 explains why that can be attractive: VXLAN carries a virtual Layer 2 network across a Layer 3 underlay, while conventional VLAN identifiers top out at 4094. The RFC also describes spanning tree leaving links idle in large virtualized networks. A tunnel does not abolish Layer 2 at its edges; it changes where the physical network must carry the burden.

His second target was spanning tree itself. A routed network can expose topology and failure through routing state instead of an opaque, growing broadcast domain. Stockebrand said proper dynamic routing could offer usable convergence and better scale in the cases he had in mind. The audience heard no public incident log or before-and-after timing. His unnamed customer remains an example from practice, not a verified industry trend.

There is a useful tension in the room: putting more IPv6 into a network only helps if it lets something else come out. Stockebrand's close was an instruction to delete unnecessary mechanisms from the design. That is an operator's wager, with hardware licences, staff skills and application dependencies on the other side of the table. The payoff has to appear in the running network, not in the slide title.