Chapter 08 - Networking: Diffusion 1972-1979
8.2 Packet Radio and Robert Kahn: 1972-1974
Shortly after the ICCC, the indefatigable Robert Kahn left BBN for DARPA to begin planning the world’s first packet radio network. The consummate engineer, he had the experiences of both Arpanet and ALOHAnet to inform his judgment. As Kahn notes:
The ALOHA system was to packet radio like the original timesharing computer was to Arpanet.
Kahn soon convened an informal working group, including Vint Cerf and Robert Metcalfe, to stimulate his thinking and engage their interests. Two challenges had to be met if they were to interconnect the packet radio network with the Arpanet. First, known problems with the communications protocol of Arpanet - the Network Control Protocol, or NCP - had to be solved. Second, a means to interface a packet radio network to Arpanet had to be conceived.
The most glaring problem with NCP was the lack of end-to-end, or Host-to-Host, error-correction. In the Arpanet, NCP error correction was provided for packet routing between the IMPs but not for messages passed between the Hosts and the IMPs. Yet in operation, errors did occur and thus no end-to-end reliability in Arpanet existed. The probability of errors from a roaming packet radio terminal would be significantly greater than the hard-wired terminals of Arpanet. In a radio network, signals could be affected by static in the air or other radio signals, or be disrupted when communicating terminals entered tunnels or buildings. There were other functions subsumed within NCP that also needed to be elevated to the Host level such as flow control and packet segmentation. So NCP needed to be redesigned both for a better functioning Arpanet as well as to support interconnection of a radio network.
The second challenge was how to interconnect the comparatively slow Arpanet to a fast packet radio network. Arpanet was designed assuming link speeds of 50 kilobits per second (kbps). A packet radio network would operate at 100 to 400 kbps. At first the mismatch in operating speeds seemed to have an easy solution: create a gateway between the two networks that converted the messages from one network into acceptable messages for the other network. A gateway had its problems, however, especially if there were many gateways/networks and terminals roved from one to another. NCP assumed there was only one network much less gateways or the even more complicated situation of a terminal sending a message to one gateway and expecting an answer on another, potentially not pre-determinable, gateway. So NCP had to be re-thought.
In June 1973, at the AFIPS NCC conference in New York City, Kahn chaired a meeting to discuss the Host-to-Host protocol issues.18 Following the meeting, Kahn and Cerf began the uncertain task of redesigning NCP.
ARPANET Completion Report, p. III-72