Hf is probably a special case. We would like to use HDLC to the extent possible. But, link reliability would be poor. We need a combination of ARQ and FEC (forward error control) protocol. We envisage a special board (similar to the TNC) for hf packet links. This subject will be developed later. Clues to our thinking are in the hf standards section of Paul Rinaldo's paper in Volume 1 of the proceedings of the ARRL Amateur Radio Computer Networking Conference, October 1981 a nd in AMRAD VuGraphs 82-31, "Proposed Packet HF Channels," and 82-32, "Rx for HF Packets."
Network Level
The network layer has often been divided into two sublayers, called the "network" sublayer, and the "internet" sublayer. The first is responsible for maintaining control over a group of links, while the internet layer is used to send data back and forth between these individual networks.
Network Sublayer
The network sublayer (called by NBS layer 3A) is responsible for maintaining the individual link connections from the link layer, a nd it is also responsible for routing the data from the sender to the proper receiver(s). Since there can be several link level connections going on at once, there is the potential of overloading the network with data. A system similar to level 2's flow control is used at the network level to make sure the network doesn't get overloaded.
The network layer is also where group connections can be made.
After a lot of research, and talking to several people in the networking field, we recommend that the amateur packet community use the X.25 virtual circuit standard at the network sublayer of layer 3. Usually the individual networks will be reliable enough to support a virtual circuit system, such as X.25. We do not envisage use of the datagram provision of X.25; no one else is using it either. (X.75 would apply to the AMICON. For hf, we feel it unlikely that a network sublayer protocol is either needed or practical. This will be left to separate study.)
Again, we recommend static addresses. The X-25 format calls for 14 quartets (semi-octets), BCD encoded with A-F not used in the current version of X.25. That forces us to encode call signs in their decimal equivalents; i.e., 2 decimal digits per letter in the call sign. (This is something that could change in X.25, opening up a future possibility of something a bit more elegant here.)
Internet Sublayer