Sebastian Moeller <moeller0@gmx.de> wrote:
    >> sadly, they aren't doing IP processing.  I don't think that they ever
    >> will decide to for NIH reasons.  I suspect that their SDN hardware
    >> probably can, and I think that SR6 is probably ideal for their use,
    >> but...

    > 	Why would SR6 be better than any other "underlay" encapsulation here,
    > like MPLS or SR-MPLS? Naively put this seems to trade 4 byte per label
    > with a full 40 byte IPv6 header plus 8 bytes plus 16 bytes per

SR6 collapses all those things into a single IPv6 forwarding engine.
The idea is that it's converged.   As for the 44 byte header, as envisioned
by the SR6 people, they would use the original header, but that's against the
Ipv6 architecture.  There are ways to compress that header down if you need
to do that.

    > "label". Asking out of genuine curiosity what are those additional
    > bytes in overhead actually buying (except the freedom from MPLS).

IPv6 all the way down.