[ax25-layer2] Still lost

[J. Lance Cotton]

pete at ae5pl.net wrote:
> On the contrary, sparsely populated areas will be better served since
> their physically wide-area LANs (as opposed to large multihop LANs) will
> be interconnected to other areas via layer 3.  One of the most important
> advantages to this method is that each LAN will be isolated from the
> layer 2 overhead (such as discovery) encountered on the other LANs.

A good way to imagine an existing example is with the APRS-IS integration
and APRS messaging, though not necessarily a "clean" implementation.

You can send an APRS message addressed to "STATION-2" out to your local
144.39 MHz LAN. It gets digipeated once (for whatever reason). An I-Gate
hears it, sends it on to the APRS-IS.

[A more router-like I-Gate would only send it on *IF* "STATION-2" hasn't
been heard by the I-Gate's receiver or some similar action. This would be
less APRS-like and more generic Layer3-like.]

The APRS-IS is a separate collision-domain with a much larger bandwidth than
the 144.39 MHz 1200baud LAN. It's a "backbone".

All the other I-Gates are listening to this backbone. One of the I-Gates has
heard "STATION-2" locally. It captures the message packet off of APRS-IS and
sends it out on the local LAN, where "STATION-2" receives it and acts
appropriately.

Know that APRS-IS is not the optimal way -- the IS is essentially one very
large collision domain. Now think of the Tier-2 APRS-IS servers. Imagine if
the I-Gate near STATION-1 was connected to server1.aprs2.net. Imagine that
the I-Gate near STATION-2 was also connected to the same server. Imagine if
that Tier-2 server "knew" this and didn't pass that message packet up to the
main server, but just passed it back out to all of the other items connected
to itself. [Or even better, only to the I-Gate that heard STATION-2.]

This cannot be done on the APRS-IS right now because it would require each
I-Gate to tell the server what it's "local" members are and/or require the
server to remember all this. But the concept is one of separating collision
domains and passing knowledge around about where things are. But you all
know this :-)

In the case describe above, the digipeater has no action to take on the
Layer3 level. If STATION-1 were near enough to the I-Gate, the digipeater
would not be necessary.

Analogy to Ethernet:
* STATION-1 is a NIC+Computer
* the digipeater is a hub [maybe needed, maybe not]
* The I-Gates are routers [aka gateway -- two ports, connecting (or
separating) two collision domains.

Problems to overcome:
* How do the routers, gateways, igates [insert favorite term here] find out
about other routers? Clearly the APRS-IS method is non-optimal. It limits
networks to two "levels" -- local and "the backbone". The only reason this
works well is because "the backbone" is globally accessible.

Answer: Look to the Internet -- Routers have auto-discovery.
New Problem: This doesn't work well on Ham stuff -- it's too transient at times.

New Problem: It takes two to tango, but a three-way tango sure does look
funny. Sure College Station, TX and Temple, TX could agree to set up a
backbone link between their LANs. What if Dallas, TX wants to join in? No
problem -- another router and it's done. What if Miami, FL wants to join in?
Gotta find a way to get there. Does everyone in between agree on your
discovery protocol?

I suppose that's one reason for this SIG -- agree on a method. Cool :-)

I don't understand enough about routing protocols and discovery methods.
What works for the Internet (and works pretty well!) is hard to do with Hams
because of the smaller and sparser infrastructure (including system admins
of LANs who will set up the equipment)

Any method chosen would do good to not require a large infrastructure, but
not prohibit it, either. I think what's being talked about here is good --
the actual form of the "backbone" links are not important as long as they
exist. It could be phone, Internet, 19.2k RF link, etc, just as long as a
router is able to talk to two or more of them.

All errors above are mine -- I claim them 100% and accept any corrections!

-- 
J. Lance Cotton, KJ5O
joe at lightningflash.net
http://kj5o.lightningflash.net
Three Step Plan: 1. Take over the world. 2. Get a lot of cookies. 3. Eat the 
cookies.