On donderdag, aug 7, 2003, at 00:40 Europe/Amsterdam, Vernon Schryver
wrote:
I look forward to seeing the IEEE reinvent the network layer and put
it
_below_ the link layer. This should be fun.
How is the IEEE reinventing the network layer this time? They really
have tried in some previous efforts, but I don't see that this time.
I think they have to in order to make this work transparently to
everything above the link layer.
(What else than a network layer would you call something that resolves
"end-to-end" addresses into next hop addresses, which is necessary to
navigate across the mesh.
What end-to-end addresses in the IP sense are you talking about?
If you have a link layer with a bunch of nodes where the reachability
between some nodes is provided by intermediate nodes, the MAC address
starts to exhibit end-to-end properties. (Hm, I guess we have this in
ethernet switching today as well.)
This would also need some sort of routing
protocol and a hop limit field.)
I agree they'll need some sort of routing protocol, but despite my
religious affliation with the Churches of RIP and IP, I do not see
that a "hop limit field" is a requirement. They will need to prevent
or deal with loops, but IP-style decrementing TTL or RIP-style
incrementing metric fields are not the only way. For example, perhaps
they could make timestamps work. Or perhaps they can somehow guarantee
that loops are impossible.
HL is the cheapest way to do it. Try running spanning tree (which bends
over backwards to avoid temporary loops) in a non-trivial topology and
you'll agree.
Do you really think that the rest a path through the Internet would
benefit from the route flapping and apparently random decrementing
of the IP TTL field if they were to use the IP TTL field to deal
with loops?
I certainly wouldn't presume the answer is "no" although it might be.
Regardless of the technical issues, do you really think that only the
IETF is allowed to think about routing protocols?
Of course not. All the best IGPs were invented outside the IETF. :-)
Obviously having wireless mesh nodes route IP would be much too
simple.
That statement is not obvious to me, except in standards committee
turf war terms. My intuition does suggest that none of RIP, IGRP,
EGP, BGP, HELO, or any other IP routing protocol would work well for
what they're trying to do.
Besides, could it be that they want to carry data that don't look
like IP packets? Do you think the IETF should outlaw such heresies?
There are no easy answers here. The only thing I know is that the
number of supposedly available bits on the medium not used for payload
is steadily increasing.