FDDI-Ether bridges

[mogul (Jeffrey Mogul)]

Arggh.  Let's recognize this for what it is: a data-link layer problem.
IP MTU Discovery is (and I'm putting on my chairperson's hat) a network
layer issue.  How the individual hosts and gateways determine the MTU
of the subnets to which they are connected should not be solved by
an IP-layer protocol

Ok, so maybe we have to think about this problem anyway.  I don't suppose
that it's productive to say that people who try to bridge Ether
and FDDI are brain-dead, so I won't.

Let's suppose that FDDI uses something like ARP (I don't know enough
about FDDI to verify this).  What we need is for the ARP entry for
each local (same-subnet) host to be marked with the usable MTU for
that hop.  For Ethernet hosts, that's easy: it's always 1500, even
if the other end is really FDDI.  And (as if by magic) you don't have to
change a single byte of code, since Ethernet hosts already assume
an Ethernet MTU (duh).

OK, so the FDDI hosts have to do something smarter.  Perhaps we need to
run a modified ARP on the FDDI side that communicates the data-link
layer MTU along with the addressing info.  Call this ARP+.

Here is where the translucent bridges come in.  Whenever they see
an ARP+ packet goes between an Ethernet and an FDDI net, the bridge
has to translate the protocols and (on the FDDI side) supply an
MTU of 1500.  This solution doesn't generalize if there are more
than two MTUs in the bridged LAN, unless we change everyone to
run ARP+ and use the translucent bridges to adjust the MTUs in
the ARP+ packets as necessary.

Have I convinced people that the FDDI-Ethernet issue is separable?

-Jeff