*Keep this in mind if you answer this message.* *Please see this process on how to deal with suspicious emails.* · This message originates from outside our organisation, either from an external partner or the internet.

> On Jun 6, 2012, at 3:16 PM, Charles E. Perkins
wrote:

>> On 6/6/2012 12:00 PM, Fred Baker wrote:

>>> That said, to me, the data link layer tells me how to layer
message exchange onto the physical layer and as a result get a message
from one
>>> end to another of a single physical medium.
When I start concatenating physical connectivity and have routing questions

>> 

>> Could they be bridging questions?

>

>here you hark back to a debate that actually kinda-sorta made sense
20 years ago, but makes no sense in this context. IEEE 802.1, and specifically
>802.1d, is a routing technology. It doesn't route in the sense that
OSPF or BGP do, but it figures out how to get packets across a network
from >point A to point B.

Actually, this debate makes a lot of sense today.
 In fact, this is the way that a lot of companies have been developing
& fielding multihop, mobile ad hoc networks for over 4 decades now
- going all the way back to the original DARPA funded Packet Radio work
in 1970s as part of the original Internet project.  Namely multihop
routing below IP, so that the multihop wireless network below IP looks
like a 1 hop IP link no matter how many wireless hops it takes.  In
other words, the wireless network does routing analogous to ethernet switching
(routing) below IP.
And we are still going strong this way today with
over a dozen different MANET wireless technologies today that are working
with multihop routing below IP - both from my company and other companies.
 I believe that in total, there are over 100,000 fielded wireless
nodes of one type or another of these MANET technologies. Some of these
wireless technologies with MANET routing below IP that have been described
in public domain documentation include Mil-Std-188-220, MARLIN (STANAG
4691), Link 22 (although the final fielded standard dropped support for
IP at layer 3), EPLRS, and FlexNet.  Some of these waveforms call
this sub-layer below IP the Intranet, others the wireless Subnet, others
the Data Link.  
As an aside, you may ask why are there so many different
types of wireless technologies?  Well the answer is that there are
lots of different requirements with different RF channel conditions with
different fading conditions, frequency allocations (different freq bands
and bandwidths), data rates (from burst rates with kbps to 100s of Mbps),
threats (jamming or not jamming), user traffic with different QOS requirements
(from ETE latencies of 10s of milliseconds to minutes, ETE message successful
delivery rates (from 90% to 99.999% success), traffic mixes (unicast only,
broadcast only, to mix of uni-, multi- & broadcast), etc.  

>>> , I'm in the network layer; the question is "where".

>> 

>> In my understanding, there isn't any ambiguity about this because,
in

>> the network layer, network addresses are used.  In layer
2, layer 2 addresses

>> are used (e.g., for bridging).

>

>again, harking back to a discussion that is kind of irrelevant. Are
you going to tell me that an IP address used by two layers, one of which
is >doing subnet routing using, say, OSPF, and the other is doing host
routing through AODV, are not both in the network layer? If so, then why
would a >network doing two layers of routing, one of which is using
subnet routing using, say, OSPF, and the other is doing host routing through
802.1d, is >not both in the network layer?

The reason all of the waveforms that I mentioned do
routing below IP is that they are all omnidirectional wireless and multihop
networks (without base stations or access points with out-of-band infrastructure
to use for routing).  Thus, MANET operation for these networks is
much more than just routing.  It includes determining what is a good
neighbor and under what conditions (e.g. predicted transmission success
probability for given FEC, transmit power, frequency, antenna setting,
etc). It includes figuring out which neighbors to address for any given
transmission as well as accounting for the self-interference caused by
the transmissions. It also includes allocating more resources (e.g. time
slots) to adapt to QOS and traffic loading at different nodes in an RF
neighborhood. In general, then MANET in these networks means cross-layer
optimization of the physical modulation, the channel access, and the Intranet
routing. (For example, see my IETF MANET 12 Jul 2002 13:18:43 email, and
follow on discussion threads, on "Interference in RF networks.")
 In general, I claim that since this cross layer optimization cannot
be performed by IP MANET routing alone, it makes architectural sense to
do MANET Intranet routing below IP and just provide an abstracted IP link
cost to the IP routing for that wireless MANET network technology.
As an aside, IMO most work in the IETF MANET has been
concerned with 802.11 networks and cross-layer optimization have focused
on its CTS/RTS MAC layer.  This is why work was done to allow jitter
to be inserted by the MANET protocols as a type of cross-layer optimization
for 802.11 as an Internet Draft circa 2007 and included in RFC 6130 NHDP
in 2011. Well Packet Radio already had that optimization included back
in 1970s and in Mil-Std-188-220 in 1990s.
So, do I think that the IETF MANET work makes sense?
Yes! For example, some of the wireless Intranet technologies I mentioned
above are using optimized variants of IETF MANET protocols, namely OLSR.
The most common optimization is to eliminate the use of Hellos for neighbor
discovery (because it is part of the MAC function) as well as using the
local MAC layer information to route to nodes that may be 1 or 2 hops away.
However, I think that some of the IETF MANET work
is not applicable to these Intranet MANETs. For example, Hello messages
should be optional if the MAC layer can provide that information directly.
In fact, trying to determine neighbors from just hellos alone can lead
to poor performance in many of these networks because it doesn't take into
account interference as an example and doesn't help select which good nodes
to use as good neighbors.
Twenty years ago, I used to say that wireless MANET
routing **must** be below IP to enable such cross layer optimization. I
no longer claim that because I now see some wireless technologies where
IP layer MANET routing alone works just fine.  However, I am now concerned
when people claim that wireless MANET routing must be at IP layer only
- because they don't know or understand the issues involved with the cross
layer optimization I discussed above.  And, it confuses non-networking
people who say "if the IETF MANET says it, then it must be true."

>>> the Internet layer is an upper sub-layer of the Network Layer,
and Ethernet/ATM/IP Tunnels/IP re-used within a manet|roll domain and etc
>>> are in a lower sub-layer I call the "Intranet"
sub-layer. But yes, I think we agree that the LAN, however comprised, is
in a sub-layer 
>>> lower than the Internet layer. 

>> 

>> Do you envision that an "Intranet" layer would use Internet
addresses at all?

>

>RPL is a pretty good example of a routing protocol that makes that
assumption. If AODV is used and the node ID is an IP address - very reasonable
- >sure.

I have seen multiple different types of addressing
used in the Intranet below the IP layer.  In most cases, the wireless
Intranet has its own address separate from the IP address.  In many
cases, this Intranet address maps to an IP subnet address (say the lower
octect or two) for that wireless network.  Hence this is why some
of the above mentioned wireless MANETs call this layer below IP the Subnet
Layer rather than the Intranet layer.   In other cases, the wireless
Intranet has its own address and a ARP or ARP-like function maps the Intranet
address to the IP address. 

Jim Stevens