[c-nsp] L2 ethernet multipoint over IP?

[Peter Rathlev]

We're in the middle of deploying a large-ish (40k access ports) new
network on a hospital. To avoid problems with large broadcast domains
and large STP domains we have split the physical network into a number
of separate sections with no L2 contact between them. Termination is on
Cat65 Sup2T using subinterfaces on 10G ports in 6904 cards. We're using
HP 7500 for aggregating Nexans microswitches (4 downlink GE ports, two
uplink SFP ports) and a pair of HP 7510s together with 480-960 Nexans
switches make up one STP domain. (Broadcast domains are further reduced
in size via different VLANs.)

For some time we have been in a (not so constructive) dialogue with
certain vendors of medical equipment. Many of these are not accustomed
to actually routing things and rely heavily on direct L2 contact between
the individual devices. Some of these utilise multicast, but they are
not willing to have it routed.

Internally we have discussed wether it's possible to construct some kind
of "overlay" network that provides L2 services across an IP network. It
would need to be multipoint, and we're possibly talking > 100 ports for
some of these L2 domains.

How could one accomplish this? The perfect solution would be something
utilising multicast for discovery of member nodes and some kind of
multipoint dynamic mesh tunnelling with MAC learning and whatever else
is needed for providing L2 services. But either my google-fu is weak or
nothing like this exists.

I've been thinking of a hub-and-spoke solution using L2TPv3 but I have
difficulties imagining a hub design that actually scales to maybe dozens
of L2 domains each with possibly more than 100 spokes.

Does anything like this exist? It doesn't have to be Cisco, that's just
what we know best. Any other suggestions are very welcome.

Thanks in advance!

(PS: It has to be pure IP as seen from the rest of the network, so no
MPLS unless it's something like MPLSoGRE.)

(PPS: Bonus points for spoke/mesh member equipment with only passive
cooling, so it can be placed together with the "client" devices.)

-- 
Peter