previous | contents | next

Chapter 24 ½ The Interface Message Processor for the ARPA Computer Network 415

of each computer that might need network connection is probably economically inadvisable. Third, because of the desirability of having several Host computers at a given site connect to the network, it is both more convenient and more economic to employ IMPs than to provide all the network functions in each of the Host computers. The whole notion of a network node serving a multiplexing function for complexes of local Hosts and terminals lends further support to this conclusion. Finally, because we were led to a design having some inter-IMP dependence, we found it advantageous to have identical units at each node, rather than computers of different manufacture.

Considering the multiplexing issue directly, it now seems clear that individual network nodes will be connected to a wide variety of computer and terminal complexes. Even the initial ten-node ARPA Network includes one Host organization that has chosen to submultiplex several computers via a single Host connection to the IMP. We are now studying variants of the IMP design that address this multiplexing issue, and we also expect to cooperate with other groups (such as at the National Physical Laboratory in England) that are studying such multiplexing techniques.

The increasing interest in computer networks will bring with it an expanding interaction between computers and communication circuits. From the outset, we viewed the ARPA Network as a systems engineering problem, including the portion of the system supplied by the common carriers. Although we found the carriers to be properly concerned about circuit performance (the basic circuit performance to date has been quite satisfactory), we found it difficult to work with the carriers cooperatively on the technical details, packaging, and implementation 'of the communication circuit terminal equipment; as a result, the present physical installations of circuit terminal equipment are at best inelegant and inconvenient. In the longer run, for reasons of economy, performance, and reliability, circuit terminal equipment probably should be integrated more closely with computer input/output equipment. If the carriers are unable to participate conveniently in such integrations, we would expect further growth of a competing circuit terminal equipment industry, and more prevalent common carrier provision of bare circuits.

Another aspect of network growth and development is the requirement to connect different rate communication circuits to IMP-like devices as a function of the particular application. In our own IMP design, although there are limitations on total through-put, the IMP can be connected to carrier circuits of any bit rate up to about 250 kilobits; similarly, the interface to a Host computer can operate over a wide range of bit rates. We feel that this flexibility is very important because the economics of carrier offerings, as well as the user requirements, are subject to surprisingly rapid change; even within the time period of the present implementation, we have experienced such changes.

At this point, we would like to discuss certain aspects of the implementation effort. This project required the design, development, and installation of a very complex device in a rather short time scale. The difficulty in producing a complex system is highly dependent upon the number of people who are simultaneously involved. Small groups can achieve complex optimizations of timing, storage, and hardware/software interaction, whereas larger groups can seldom achieve such optimizations on a reasonable time scale. We chose to operate with a very small group of highly talented people. For example, all software, including software tools for assembly, editing, debugging, and equipment testing as well as the main operational program, involved effort by no more than four people at any time. Since so many computer system projects involve much larger groups, we feel it is worth calling attention to this approach.

Turning to the future, we plan to work with the ARPA Network project along several technical directions: (1) the experimental operation of the network and any modifications required to tune its performance; (2) experimental operation of the network with higher bandwidth circuits; e.g., 230.4 kilobits; (3) a review of IMP variants that might perform multiplexing functions; (4) consideration of techniques for designing more economical and/or more powerful IMPs; and (5) participation with the Host organizations in the very sizeable problem of developing techniques and protocols for the effective utilization of the network.

On a more global level, we anticipate an explosive growth of message switched computer networks, not just for the interactive pooling of resources, but for the simple conveniences and economies to be obtained for many classes of digital data communication. We believe that the capabilities inherent in the design of even the present subnet have broad application to other data communication problems of government and private industry.



Baran [1964]; Baran, Boehm, and Smith [1964]; Boehm and Mobley [1966]; BBN Report No. 1763 [1969]; BBN Report No. 1822 [19691; Brown, Miller, and Keenan [1967]; Carr, Crocker, and Cerf [1970]; Cuadra [1968]; Davies [1968a]; Davies [1968b]; Davies, Bartlett, Scantlebury, and Wilkinson [1967]; EDUCOM EIN Catalog; Everett, Zraket, and Benington [1957]; FCC [1966a]; Ford and Fulkerson [1962]; Frank, Frisch, and Chou [1970]; James [1966]; Kaplan [1968]; Kleinrock [1964]; Kleinrock [1969]; Kleinrock [1970]; Marill [1966]; Marill and Roberts [1966]; National Library of Medicine [1968]; NOC Symp. [1968]; NOC Symp. [1969]; Perry and Plugge [1961]; Roberts [1967]; Roberts [1968]; Roberts [1969]; Roberts and Wessler [1970]; Scantlebury, Wilkinson, and Bartlett [1968]; Steiglitz, Weiner, and Kleitman [1969]; Sung and Woodford [1969]; Teitelman and Kahn [1969].

previous | contents | next