Why does the operator of a layer 2 exchange care (or know) what
protocols your are using? IPv4, IPv6, heck I remember seeing
Appletalk, OSI and DECNET on MAE-EAST. What consenting network
operators do....

What consenting network operators do bilaterally in an L2 environment
where their actions might possibly affect other customers of the L2
exchange is of great interest to the exchange point provider.

Take multicast at an exchange point, for instance. IGMP is an edge
protocol, not used in the exchange of multicast information between
autonomous systems. Since there's no IGMP to snoop, multicast packets
are flooded to all ports of an Ethernet-based L2 fabric. Combine that
fact with the fact that a range of port speeds are offered, and stir:
you could get into a sitation where multicast traffic solicited by one
high-speed port customer of another high-speed port customer "drowns
out" lower-speed port customers (especially when available speeds span
two orders of magnitude).

For this reason, multicast traffic exchange on Ethernet-based L2
exchange points is often conducted on a separate switch fabric, or a
separate VLAN (as in the case of PAIX). When we see PIM on the main
"unicast" VLAN, we ask the PIM speaker to take it to the multicast
VLAN (for which we provide IPv4 address assignment just like the

"Consenting network operators" also engage in practices such as
connecting their exchange point switch ports into an aggregation
switch of their own, and then connecting their aggregation switches
together to implement private peering. If not properly configured and
maintained, this has the potential to introduce loops in the switching
fabric that can lead to a variety of failure modes for other

Engineering an L2 fabric that implements an administrative boundary
between many networks prsents unique challenges, from cases such as
those outlined above, helping providers learn how to detect and
correct cases having default pointed at them, all the way to
explaining to Juniper router owners that their "policed discards"
counter increments because other participants emit CDP packets (and
some of the other non-IP protocols you mention, but most often CDP).

More germane to your questioning of Bill's point, though, items such
as IPv6 address assignment, and an engineering staff prepared to deal
with IPv6-related questions and issues, determine whether an exchange
point supports IPv6 above and beyond just letting it happen on their
switch fabric.

(now VP of Engineering at PAIX, as well as being a founder, etc.)