[ax25-layer2] AX.25 and Layering

[Timothy J. Salo]

I haven't thought about this for a while, but a few quick comments about
AX.25 and protocol layering follow.

Perhaps my most fundamental complaint about the design of the AX.25
protocol is that is munges together, or includes functionality
from, the link (layer 2), network (layer 3), and transport (layer 4)
layers.

o	The HDLC framing is clearly link-layer a function
o	The source-routing capability is clearly a network-layer
	function
o	The end-to-end acknowledgment/retransmission function
	is a transport layer function.

Having said that, within broad limits, I wouldn't get overly religious
about what functions ought to occur at what protocol layers.  For example,
there are link-layer protocols that perform hop-by-hop retransmissions,
(although they interact poorly with TCP).  There is also a view that
the link layer should perform routing, traditionally viewed as
a network layer network layer function.  I think the new sensor
network protocol being developed by the IEEE will do routing at
essentially the link layer (which permits any network layer to be
transported).  Note that I generally don't agree with the view
that the link layer should route, but there might be cases where
it is appropriate.  In short, I think it is more important to understand
_why_ you do or don't want to perform a particular function at
a certain layer, than it is to do it just like everyone else.  This is
particularly true for wireless networks, which are still an area
of active research.

If you are designing a new solution, picking an addressing strategy
is pretty fundamental.  Do you want a flat address space, where
addresses are assigned without regard to network topology (e.g.,
IEEE MAC addresses and call signs), or do you want some sort of
hierarchical address assignment, where addresses that are similar
are topologically close to each other (e.g., IP, where addresses
on a subnet all have the same high-order bits).  The choice of
addressing strategy has significant implications for your routing
strategy, particularly if you want to do any sort of dynamic routing.

(By the way, some of the solutions for using IP in mobile, wireless
networks essentially use a flat address space within the wireless
network.  That is, IP addresses within the wireless network
have no topological significance -- nodes just use whatever
IP address they show up with.)

If you are going to develop a new link-layer protocol, I would
look at, among other things, the protocols used for space
communications.  Note that I said "look at", not necessarily "use".
The space communications people appear to have historically been
isolated from the rest of the networking world, and apparently thought
they had to have unique solutions.  Nonetheless, their protocols
embody a good understanding of and lots of technology tailored
for wireless networks.  See, for example, http://www.ccsds.org/.

-tjs