This is a temporary index for a collection of papers about packet-switching in the 1970s. I have deliberately excluded papers on ARPANET and CCITT X25 on the grounds that these systems have been adequately covered elsewhere. I have tried to choose papers which describe wide-area networks which were actually built and operated. I have excluded DECNET (aka DNA) on the grounds that it lacked a transport capability in 1980.
I may include papers on the Japanese DDX-1 system. I am still looking for technical papers on the following systems:
United Telecom before 1980;
Scannet from Norway
Centernet from Denmark
These papers date from the 70s when there were no standards for packet switching. For reasons explains in The Evolution of Packet Switching it was relatively easy to implement a network technology and provide better customer service at lower cost than with other technologies of the era. Proponents of both circuit switching and packet switching were active. Within the packet switching camp, there were sharp divisions between believers in datagrams and virtual circuits. ARPANET and CYCLADES are the most prominent examples of datagram technologies. RCP, TYMNET and X.25 are the best known virtual circuit systems.
This index begins with a group of survey papers which mention multiple networks. It is followed by papers describing specific networks. Most of these single network papers were written by people who were closely associated with the design or implementation of the subject network. The networks are listed in alphabetical order.
Lawrence Roberts was deeply involved in the creation of ARPANET due to his position as head of computer research at ARPA. He lists research efforts which led to the creation of ARPANET. Several other early packet-switching systems are briefly described. The economic and regulatory reasons which caused several other organizations to re-implement packet-switching are given. Prescient remarks about the future conclude the paper.
This survey is based mainly on published papers from ICCC 74 supplemented by discussions at ICCC 74. ARPANET, CYCLADES, CTNE's RETD, Nippon PTT's DDX-1, EIN, EPSS, NPL, and RCP are covered. Services and uses are mentioned with few details on interfaces or internal workings of the various networks.
Peter Kirstein described many planned and actual European networks. Some of these were never implemented. Short but useful descriptions of Tymnet and GEISCO network are included.
Schwartz et al describe four computing systems which are designed to service remote terminals (individual or batch). Links to leap into the sections of the paper are: TYMNET, GEISCO, NASDAQ, and CSC INFONET.
Louis Pouzin is the principal architect of CYCLADES. Flow control goals and possible solutions are discussed. The flow control schemes of several specific networks are mentioned. M. Pouzin understood the compleixty of the problem. Observe that people are still debating congestion control techniques for the Internet.
This paper was written by a specialized provider of computer services. He documents some of the networking problems of the 70s.
Many papers about ARPANET are available elsewhere. www.livinginternet.com is one starting point although Google Scholar may provide more detail on specific topics.
AUTODIN was a message-switching system which served the US DOD. It lacked the ability to multiplex multiple messages on a single communications link and thus was not a packet system. I have included here as an example of the technology which preceded packet-switching. Note that the SITA network has some of the functions of AUTODIN.
Bell-Northern Research was a joint venture of Northern Telecom and Bell Canada. It was the research lab which developed DATAPAC (a very early X.25 network). This paper describes experience with a modest network which provided remote access to a single host from several cities. It was designed for multiple hosts . HDLC framing but not link control was used. Link control allowed selective retransmission of lost frames. This relatively uncommon feature is also present in ARPANET and IPSANET.
Compañía Telefónica Nacional de Españ (CTNE) began operating "Red Especial de Transmisión de Datos" (RETD) in 1971. Initial applications were terminal-host communication and terminal-terminal (message switching). Other applications were introduced or planned in the 70s. It was a two level network with the lower level occupied by terminal concentrators. These devices were similar in spirit to an X.25 PAD and connected many terminals to a forwarding node. The high level nodes were connected via redundant 48kbps lines. The organizaton was virtual call rather than datagram.
The software was developed by CTNE and its clients. Univac 418 computers were used in the original node implementation. In 1979 CTNE was planning to install a new node based on multiple 8086 processors.
Three papers written by CTNE employees are presented here:
Alarcia and Herrera provide maps of RETD at various stages of development. Two services were offered at 1974: Host to terminal connection; terminal to terminal message switching with code conversion. The terminals were often multi-drop machines which required polling to elicit input. Block formats are specified; the protocol for polling/addressing control is described in some detail. Planned future services are listed. Some graphs showing benefits of packet switching are provided. {Critics have suggested that many of these benefits derive from relocating polling responsibility from the host to the remote terminal adaptor.}
1979 statistics on network size and availability are presented. Growth in network size from 1972 to 1979 is measured by traffic voluume and terminal count. Tariff structure and customer support are mentioned. Plans for new node hardware, new services and international connections are described.
LAVANDERA describes special packets are establishment, reset annd clearing etc of virtual calls. Another family of special packets exchanged between host and the terminal adapter control polling and the transfer of messages. The use ofthese packets is described in greater detail in Alarcia and Herrera above. Interesting comments about X.25 specifications (LAP vs LAP B) can be found here. Specifications for the second generation node hardware (TESYS 5) are provided. The uunderlying HW is a collection of Intel 8086 processors.
CYCLADES was a French effort which began development in 1972. It was intended to link various computers belong to French universities, research institutes and eventually the French government. This paper discusses host organization rather than the switching aspects of the network. The emphasis was on host to host communication with terminals connected via a specialized type of host.
The 1973 paper was presented about a year before CYCLADES began operation. Division of labour between hosts and the transport network is discussed. Network facilities visible at the host level are explained in some detail. The problems of domain names in a very large network are examined.
This second paper was written after about a year of operating experience with CYCLADES. Lessons learned and plans for future development are covered.
CIGALE is the transport and link control system which underlies CYCLADES. This paper talks about the transport layer and the evils of virtual circuits. Packet header format is presented but there is no detail on link control or routing.
EPSS was a British experimental network. It was intended to provide both host-host and terminal-host communication. Deployment was limited as X.25 technology was introduced by the Post Office a few years later. EPSS had an unusual link control protocol in which acknowledgements were injected into the middle of outgoing packets. Trunk speed was 48kbps and the nodes were built upon a Ferranti Argus 700E processor.
Network service and link control protocol are described here.
EPSS Node specifications and structure are presented here. Figures 1 & 2 may be of interest to readers of Bright & Smith above. The difficult problem of receiving EPSS acknowledgements is mentioned herein.
General Electric Information Services Company (a name first used in 1979) started selling time-sharing services in 1965. The service was provided from a local medium-sized computers in every market. This proved uneconomic and so GE developed network technology which allowed them to service their world-wide customers with a small number of large computers located in Brook Park near Cleveland, Ohio. This technology was deployed in the 1970-1972 era. The network is characterized by several layers of specialized processors rather than the universal "node" found in most packet-switching systems. Many of the facilities of a PS network are present such as redundancy, efficiency link sharing and routing. Little has been published about this pioneering effort. An overview of the GE network can be found in these two papers.
Kirstein provides a brief description and some analysis of GE's network.
TONET provides more detail on both applications and message formats but less analysis.
Hitachi designed a private network system for sale as a turnkey package to multi-national organizations. In addition to providing X.25 packet switching, message switching software was also included. Messages were buffered at the nodes adjacent to the sending and receiving terminals. Switched virtual calls were not supported, but through the use of "logical ports" an originating terminal could have a menu of pre-defined destination terminals.
I. P. Sharp Associates was a Canadian computer service vendor. They developed their own packet switching technology for internal use (much like GEISCO and TYMSHARE).
A group of mostly unpublished papers on IPSANET can be found here.
The National Physical Laboratory (NPL) is the national measurement standards laboratory for the United Kingdom. Donald Davies, Win Price and other NPL researcher have contributed to the theory of packet switching. Their inhouse system is a relatively modest single node network with 12 User Machines. This paper was written before the link to CYCLADES was established in 1975.
Scantlebury and Wilkinson have some personal recollections here.
Derek Barber's recollections of the origins of the NPL network and X.25 can be found here.
Livermore Radiation Lab was an a major research center for the Atomic Energy Commission in 1970. The OCTOPUS local network provided various services to users of powerful mainframes as the Lab. Semi-independent subnetworks supported file storage, TTY connection, RJE and other services. Packet technology was used except for the storage application.
Philips Research Laboratories designed and built a single node virtual call packet switch system for internal use. Most of the computers are byte oriented except for the central ICL 1904S which used 6-bit characters. The switching node converted between 6 and 8 characters when the 1904S was an end node of a virtual call. THe switching node sheltered some virtual machines which drove printer, plotter etc. The switching node discards invalid packets. Retransmission is done on an end to end basis rather than at the link level.
RCP was an experimental network created by the French PTT. It was used to gain experience with packet switching technology before the specification of TRANSPAC was frozen. RCP was a virtual-call network in contrast to CYCLADES which was based on datagrams. RCP emphasised terminal to host and terminal to terminal connection; CYCLADES was concerned with host-to-host communication. TRANSPAC was introduced as an X.25 network. RCP may have influenced the specification of X.25.
Three papers on RCP are presented here:
Despres discusses protocols both at the asynchronous terminal and at the synchronous host. Hardware to connect a character multiplexor to the PDP-11 Unibus is briefly described. There may be enough detail on link and packet level protocols to allow reconstruction of RCP.
Bache & Matras discuss objectives of RCP and how RCP experience influenced TRANSPAC specifications. They argue in favour of a virtual call network with fixed routing. In RCP as in TYMNET, a transmission frame may contain from several virtual calls. A single virtual call is formed by the catenation of channels where a channel is contained within a transmission link.
Bache, Guillou et al describe various schemes used to connect hosts and terminals to RCP. Most of these involved some kind of protocol converting computer. A PDP-11 Unibus card which emulates a time-division multiplexor is mentioned.
This paper describes REXPAC, the nationwide experimental packet switching data network being developed by CPqD, the research and development center of TELEBRAS, the holding company responsible for public telecommunications in Brazil. REXPAC uses internationally-standardized CCITT interfaces, following CPqD interpretations for X.3, X.25, X.28, X.29 and X.75.
The first experimental packet-switched Nordic telecommunication network SCANNET was implemented in Nordic technical libraries in 70's, and it included first Nordic electronic journal Extemplo. Libraries were also among first ones in universities to accommodate microcomputers for public use in early 80's.
The preceding paragraph taken from: http://edoc.hu-berlin.de/conferences/eunis2001/e/Haarala/HTML/haarala-ch2.html
is the only information which I have on SCANNET. I believe it was programmed at the University of Oslo.
SITA is a co-operative operated by the world's airlines. Its purpose is to provide communication services to its members. The network provided both message-switching (terminal-terminal) services and query/response service to connect a terminal and a host. Deployment began in 1969 at about the same time as ARPANET deployment. In 1973 the three node types were: Philips DS 714 MK II, UNIVAC 418 II, UNIVAC 418 III. Maximum link speed was 9600bps.
Systems Network Architecture was IBM's packet-switching offering. It began as to terminal-to-host network and evolved into a general network. Some problems such as routing, network control, etc. are implemented in a fashion which is somewhat different from other networks. This paper provides a good survey of SNA capabilities and how they expanded over six years.
Tymshare was an early (1964) vendor of remote computer services. Their customers tended to use 300 baud or slower full-duplex ASCII terminals. TYMNET was designed and built by Tymshare to economically provide a high quality connection of many remote terminals to several time-sharing host computers. Like several other terminal-oriented networks it was focused on virtual calls rather than datagrams. As in the later RCP network, a transmission frame usually contained data for multiple virtual calls.
This paper describes TYMNET hardware and gives details of routing and call establishment within TYMNET. Arguments in favour of virtual call organization are presented.
Beere and Sullivan discuss the problems which TYMNET was intended to solve. Some of the mistakes made before the network reached the state described in Tyme's 1971 paper are mentioned.
1972 TYMNET—A Serendipitous Evolution
by MAX P. BEERE and NEIL C. SULLIVAN
This paper was originally published in IEEE Transactions on Communications, vol COM-20, pp 511-15, June 1972.
This paper discusses TYMNET solutions to problems in flow control and route establishment. Flow control problems within datagram network are mentioned. The limitations of two central control schemes which have been used in TYMNET are analyzed.
UNINETT is a Norwegian national packet switched network established through a joint effort between the Norwegian Universities, some research institutions and the Norwegian Telecommunication administration. The original network was based on X.25; the present network uses Internet technology. This paper discusses higher level protocols and says little about the protocol levels implemented via X.25.
X.25 was developed by an international consortium of telephone companies. Bell Northern Research was one participant. Their X.25 implementation was the basis of DATAPAC in Canada. It was also sold as a turnkey system to some other PTTs such as Deutsche Bundespost. ICCC 76 contains several papers on DATAPAC which may be of interest.
Editing Policy: I have tried to preserve the appearance of the original article. In many cases this meant using a Courier or sans-serif font to imitate the original typewritten manuscript. I have attempted to copy the intention and spacing style of the originals. Some changes have been made to the original copy. In most cases these are emendations which should have been made before the paper was published (grammar & spelling errors). For one paper, I have made many minor "improvements" to the original grammar. The author was not fluent in English and his meaning was sometimes not immediately obvious to the reader. I have attempted to mark those changes by a font change.
Many drawings have been retraced to eliminate printing and scanning junk. Some drawings have slightly re-arranged to fit comfortably in the browser window. For some drawings the font size on labels had to be increased for legibility which forced a slight increase in the size of the enclosing rectangle. Some graphs and maps have had colour added. Colour is usually a replacement for dashed lines in the original. This change was reserved for those drawings where I thought colour would improve readability.
Bibliographies have heavily edited. Abbreviations in many titles have been expanded to make it easier to find the original volume. Page numbers etc have sometimes been emended. In a few cases, a hyperlink to a site which has the full text of an article has been added. (I would like to add more of these links.) The following acronyms may remain in the bibliographies:
ACM Association for Computing Machinery
AFIPS American Federation of Information Processing Societies
ICC International Communications Conference (an annual IEEE event)
ICCC International Computer Conference on Communications (a biannual event)
IEEE Institute of Electrical and Electronic Engineers
IFIP International Federation for Information Processing
IRIA Institut de Recherche d'lnformatique et d'Automatique
SJCC Spring Joint Computer Conference (an annual AFIPS event; also FJCC)
Comments may be directed to the webmaster of this site.
Roger D. Moore August 2006
Research assistance was provided by John Thompson