This paper was presented at Proc. of Symp. on Computer Networks (1975), 1--7; (in Computer Networks: A tutorial, Abrams. M., Blanc, R. P., and Cotton, I., (Eds.), (1980)).
David C. Wood
The MITRE Corporation
McLean, Virginia
This paper surveys eight packet switching computer networks with emphasis on their capabilities and the economic analyses which justify their use. The networks are characterized from the viewpoint of a potential user who might wish to access the network from a computer or terminal. The user-oriented capabilities provided by each network are identified; requirements for interfacing a user's computer or terminal to the network are shown; and trade-offs of using the network are discussed. The eight networks surveyed are ARPANET, CIGALE/CYCLADES, CTNE, DDX-1, EIN, EPSS, NPL, and RCP.
Packet switching computer networks have evolved from only two networks in existence three years ago to a situation where such networks are under development in many major industrialized countries. Moreover, packet switching networks have transitioned in this period from purely research environments to being planned as public data communications services in a number of countries. Most network developments have concentrated on implementing and advancing the technology of packet switched data communications. Little attention has been paid, at least publicly, to identifying the user-oriented functions provided by the packet switching networks and performing economic analyses to demonstrate the networks' cost-effectiveness for these functions.
This paper surveys eight packet switching networks, identifying the capabilities provided by each and reporting the economic analyses, if any, which show the networks' cost-effectiveness. The networks are characterized from the viewpoint of a potential user who might wish to access the network from a computer or terminal. The user-oriented capabilities provided by the network are identified; requirements for interfacing the user's computer or terminal to the network are shown; and trade-offs of using the network are discussed. This survey is not concerned with internal characteristics of the networks, such as routing strategies and packet formats.
The eight networks surveyed are:
| ARPANET | the Defense Advanced Research Projects Agency network in the U.S. |
| CIGALE/ CYCLADES | a French government-sponsored research network |
| CTNE | a network operated by the public carrier CTNE in Spain |
| DDX-l | an experimental network developed by NTT (the public carrier in Japan) |
| EIN | the European Informatics Network a joint research project among European countries |
| EPSS | the British Experimental Packet Switched Service |
| NFL | the network at the British National Physical Laboratory |
| RCP | an experimental network operated by the public carrier in France. |
Information on six of the networks is derived largely from papers on those networks presented at the Second International Conference on Computer Communication (ICCC-74) and discussions at the conference emanating from those papers. The other two networks, ARPANET and CIGALE/CYCLADES, are well described elsewhere in the literature.
The prime existing example of a packet switched network is the ARPA network which comprises a data communication subsystem and a collection of host computers which together make up a resource sharing network[1]. The packet switched local network at NFL addresses the same fundamental communication issues, but on a small scale[2]. Those two networks have been in existence for some years, and have largely inspired the numerous other packet switching networks in various stages of planning and implementation. However, these networks vary considerably in their design and the facilities they offer to the users.
The common feature of all the networks is the employment of well-formatted, relatively short message units called packets. The CCITT defines a packet as "a group of binary digits including data and call control signals (e.g., address) which is switched as a composite whole. The data, call control signals and possibly error control information are arranged in a specified format" [3]. Essentially, each addressed packet occupies a transmission channel for the duration of the transmission of the packet only. The channel is then available for use by packets being transferred between other users. Thus, channels between packet switches are shared between many users on a demand basis, and it becomes possible for a customer having a single link to his packet switch to engage in the exchange of data packets with a number of other users at the same time.
To a user, a packet switching service can be regarded basically as the acceptance and delivery of well-formatted packets. For the computer attached to the system (the host computer) a simple protocol is also required for the immediate link with the network, to deal with local flow and error control and with those control signals which are outside the stream of data packets. A packet switched system with these simple functions has been called a 'primitive packet-switch' or 'datagram' [4]. Such systems represent one end of the spectrum of packet switched systems.
In the simple system described above, a packet is launched into the network without knowing the state of the receiving device and a stream of packets might be sent which cannot be delivered. Although packets are unlikely to be delivered garbled, they may possibly be lost or duplicated in transit. The packet sequence can also be changed, on occasions, because of the variable delay due to routing and to retransmission following an error. Protocols which deal effectively with all these deficiencies can be left up to the host computers, or incorporated in the packet switching communications network. In the latter case, the network can be used to provide the logical equivalent of a circuit between computer and computer or computer and terminal. This has been called a 'virtual circuit.' A protocol is required to set up the virtual circuit between what is effectively a port on the computer and another port or terminal. In addition, a procedure is required for using the virtual circuit once it has been established.
Several important characteristics of a packet switching network influence the requirements imposed on the user considering connecting his computer or terminal to the network. The characteristics are defined here and used subsequently in the survey.
Any membership qualification for joining the network must be satisfied. Existing and planned networks can be categorized into two classes: private and public. Private networks are restricted to a closed community, such as an organization or a research community with common interests. Public networks are operated by a public carrier as a service. Private networks which are developed for research and experimentation purposes may not be cost-effective in providing services, but public networks will need to offer cost-effective services to survive.
The type of packet service provided may be of the simple datagram type or a virtual circuit, as discussed above. The simpler service is likely to impose greater demands on the user to perform functions otherwise performed in the communications network.
The mechanism for interfacing computers to the network may use standard host supported communication protocols or may be non-standard. The bandwidth across the interface (kilobits/second) and unit transferred (character, packet, multi-packet) is also of concern to the user.
The network mayor may not support the direct connection of terminals. If allowed, input/output between a terminal and the network may be in characters or packets, the latter necessitating an interface device at the terminal.
Interfacing a computer to the network may require changes in the operating system, depending on the interface characteristics and the features supported by the operating system.
Two general classes of user capabilities may be supported by the network: terminal to computer communication and computer to computer communication. Terminal to computer communication is the ability of a terminal, either on a host or directly connected to the network, to engage in conversational interaction with a computer on the network selected by the terminal user. Computer to computer communication is the ability of an executing program (or process) in one host to establish communications and exchange data with a process in another host, for purposes such as file transfer and remote job entry.
Costs to a computer installation of belonging to a packet switching network consist of initial connection expenses and recurring operating costs. Initial costs may include a special hardware interface for the host; installation of communication lines between host and packet switch; and the development of special host software. Operating costs will consist of direct use charges for the packet switching communications network probably based on the number of packets, and possibly indirect costs for maintenance of special host software and loss of host capacity because of the special host software.
Reasons for Using Packet Switching
A prospective user is likely to be attracted to packet switching because it is cheaper than alternative equivalent communications facilities and it provides capabilities not otherwise available.
For terminal to computer traffic, the choice of the most cost-effective communications system depends on a number of variables such as location of host and terminals, and traffic characteristics. Packet switching is undoubtedly a viable alternative for certain mixes of the variables. For computer to computer traffic, the principal alternative to packet switching is direct connection (dedicated circuits), which is only possible for two or three computers.
Packet switching provides a number of capabilities not usually otherwise available. The automatic error control results in improved transmission error performance. The redundancy of alternative routing means greater availability. Data rate and character set conversions permit data exchange between otherwise incompatible terminals and computers. Perhaps most important is resource sharing, i.e., the ability to access over the network different and unique facilities
Surveys of the eight packet switching networks follow. The objective in developing each network is identified, and its characteristics and capabilities are summarized from the viewpoint of a potential user, for example, how the user gets on the network and what he can use it for. The information on most of the networks has been verified with the respective network organizations. Any economic analyses performed by the network developers are reported.
A summary of the survey appears in Figure 1. The figure shows the major user-oriented characteristics of each network, as far as can be determined from available documentation. The characteristics refer to features defined in the preceding section. The item 'economics considered' in Figure 1 indicates whether documentation is known which evaluates the cost-effectiveness of the network in providing certain capabilities.
| Network | ARPANET | CIGALE/ CYCLADES | CTNE | DDX-l | EIN | EPSS | NPL | RCP |
| Country | US | France | Spain | Japan | Europe | UK | UK | France |
| Membership | Private | Private | Public | Public | Private | Public | Private | Public |
| Type of Packet Service | Virtual Circuit | Datagram | Virtual Circuit | Virtual Circuit | Datagram | Virtual Circuit | Datagram | Virtual Circuit |
| Host Interface Type | Non Standard | Binary Sync. | Serial Sync. | HDLC | HDLC | Non Standard | British Standard | Serial Sync. |
| Maximum Speed | 200Kb | 48Kb | N/A | 48Kb | 130Kb | 48Kb | N/A | 4.8Kb |
| User Input | Multi-Packet | Packet | Packet | Packet | Packet | Packet | Packet | Packet |
| Terminal Interface | Yes | No | Yes | Yes | No | Yes | Yes | Yes |
| User Input | Character | — | Packet | Character | — | Character | Character | Character |
| OS Modification | Yes | No | N/A | N/A | No | N/A | Yes | N/A |
| Terminal-Computer | Yes | Yes | Yes | Yes | N/A | Yes | Yes | Yes |
| Computer-Computer | Yes | Yes | Yes | Yes | N/A | Yes | Yes | N/A |
| Economics Considered | Yes | No | Yes | Yes | No | No | No | No |
N/A = Information Not Available
FIGURE 1
SUMMARY OF NETWORK CHARACTERISTICS
The Defense Advanced Research Projects Agency computer network (ARPANET) is the only existing packet switching network of its size [5]. Participation in ARPANET is limited to ARPA contractors in support of their work for ARPA, and military organizations for research and development purposes.
The network comprises packet switches, known as Interface Message Processors or IMPs, and a great variety of computers or hosts. Each IMP is connected to two, or possibly three or four, other IMPs, to form a distributed network, and supports one or more hosts. An augmented IMP, the Terminal Interface Processor or TIP, can additionally support terminals directly.
The network was begun in 1969 and entered an operational phase in 1971. Currently the network includes between 45 and 50 IMPs or TIPs, about equally divided, and over 50 hosts. In addition to nodes in the continental United States, there is also a node in Hawaii and two in Europe.
The IMP communications subnetwork provides for the delivery of messages of up to about 8,000 bits from one host to another host. The messages are partitioned into packets of about 1,000 bits for transmission through the network and reassembled in the correct order at the destination IMP.
Hosts near to an IMP are connected by an asynchronous bit-serial interface requiring special hardware at the host side. Messages are transmitted between host and IMP in full duplex. Hosts at greater distances can be connected to an IMP by a communications line in which case special error detecting hardware and software is required at the host side. In this case, packets are transmitted between host and IMP. The TIP can support a variety of terminal devices, transmitting characters asynchronously at speeds up to 19.2 Kbps.
Communication between hosts is conducted according to a host to host protocol, which is implemented as a Network Control Program (NCP) in each host. The NCPs control the establishment of logical connections between executing programs; exercise flow control over the connections; and construct and interpret headers at the beginning of each message.
Additional protocols have been defined and programs implemented using the NCP to provide network user capabilities. One of these, Telnet, (for Telecommunications Network), enables a terminal at one host, or a TIP, to establish a logical connection to a host elsewhere on the network, and appear as local terminal at that host. A file transfer capability enables the movement of files from one host to another host. A message service enables a terminal user to send a message to users at other hosts on the network; messages are stored in the recipient's mailbox file. The above three capabilities are implemented on practically all ARPANET hosts and used extensively. Other less widely implemented functions include remote job service, whereby jobs can be submitted from one host for execution on another; and graphics, in which a graphics terminal at one host can interact with a graphics application program on another host.
Claims as to the cost-effectiveness of the network have been made, although some of the assumptions used to derive the figures have been questioned [6,7]. Users have not yet been charged for the traffic they generate on the communications subnetwork, but this is possible in the future. The communications subnetwork is used particularly inefficiently by the numerous PDP-10 TENEX hosts on the network, which expect terminal input a character at a-time and perform echoing at the host. The number of packets on the network is also influenced by NCP implementation features, such as the flow control system [8].
The ARPANET has unquestionably achieved its objective of providing resource sharing; that is, enabling users to access distant unique resources. The network has also fostered collaborative effort among geographically separated groups of people working on common subjects.
Public packet switching services similar to ARPANET are expected to become available in the United States during 1975. Two companies, Packet Communications Inc. and Telenet Communications Corporation, have received approval from the Federal Communications Commission (FCC) to operate commercial packet switching or value added networks [9,10]. They are not included in this survey as little technical information is presently available concerning user interface specifications. However, the companies and their financial backers may be presumed to have determined that their service is financially viable. The question has been raised as to whether the attractiveness of the commercial packet switching networks is due to the inherent features of packet switching or whether it is because they provide the only means of using communications facilities in a demand responsive way under the existing communications marketing structure [11].
CYCLADES is a computer network linking 16 heterogeneous computers in universities and research centers in France with a 7-node packet switching network [12]. CYCLADES is intended to be both an operational tool for government use and a prototype for network experimentation in various areas, including distributed data bases.
CIGALE is the name of the packet switching communications network within CYCLADES [13]. CIGALE is a simple packet switch, analogous to the datagram. In particular, there is no end-to-end flow control, error control or sequencing. Those functions are included in the host-to-host protocol. In CIGALE, the host address space is independent from the network topology. Hosts may be connected to several nodes (i.e., dual homing), and several host addresses may be reached over the same line. The latter allows several logical hosts to be connected on the same line, such as distinct hos t protocols within the same host, several virtual hosts (as in IBM-CP67), or several hosts reached through a front-end computer.
Hosts are assumed to be located at some distance away from a CIGALE node, requiring connection via leased lines. The electrical interface is CCITT V24 standard up to 19.2 Kbs and V35 at higher speeds. Manufacturer-provided transmission procedures, namely, transparent binary synchronous, are used between a host and CIGALE. Consequently no modifications to host operating systems are necessary, and host to host protocols may be implemented as user programs. The ISO standard 16 bit cyclical checksum is employed. CIGALE has no terminal handling facility for attaching terminals directly. However, a terminal concentrator or mini-host has been developed to provide access to the network- to users not attached to a host. The terminal concentrator uses the MITRA-15 minicomputer which is also used for the packet switches in CIGALE.
Services available on CYCLADES at the end of 1974 were time sharing, remote batch and file transfer. The CIGALE and CYCLADES projects are primarily research oriented; particular effort is noticeable in the development of host protocols and in concerns for interconnecting networks. Development of new capabilities, such as resource sharing is a prime objective.
The costs of developing CYCLADES have been carefully recorded and broken down by activities. Applications of CYCLADES have evolved more slowly than anticipated and no cost comparisons with other communications systems have been reported.
The Compania Telefonica Nacional de Espana (CTNE), as the public carrier in Spain, has opened a public packet switching service with exchanges in Madrid and Barcelona [15]. About 500 terminals are connected to the network at present, the major user being the banking community.
Two distinct services are provided by the network: real time and message switching services. The real time service is directed to closed groups of users, in which the traffic transmitted by/towards a terminal is directed towards/from a determinate host computer, without possibility of direct traffic between the terminals. This service has a fixed address, which means that the messages generated by one terminal will always go to the same host computer.
The message switching service also serves closed groups of users. It enables an exchange of messages between any two stations (terminals or computers) or a closed group. In contrast to the real time service, each terminal can communicate with any other in the same closed group. The network adjusts the message to the condition required by the destination stations, providing code change, speed change and packet arrangement change if necessary.
Terminals are connected to the packet switches by concentrators and multiplexers. In the real time service, the messages generated by a terminal always have a fixed address, therefore the packets do not have headers between terminal and concentrator. Once this packet reaches the concentrator, a 56 bit header is added and remains until the packet reaches the host computer.
A comparison is presented of the packet switching network versus point to point circuits for banking applications [15]. Although not clear from the paper, it is understood that the greater number of terminals supported by the computer under packet switching is due to the front end computer unloading the main CPU. The redundancy in the packet switching system results in less time lost due to failures than with the point to point systems.
A system integrating circuit switching and packet switching has been developed by Nippon Telegraph and Telephone (NTT) in Japan [16]. An experimental switching system DDX-1 (Dendenkosha Data Exchange) has been designed and implemented; DDX-1 uses time division multiplexing to dynamically share the wide band line capacity between circuit switching and packet switching. An earlier paper, presented at the Third Data Communications Symposium, analysed the traffic handling capacity of packet switched and circuit switched networks [17]. That paper concluded that the cross-over point occurs at about 5000 bits per call, with circuit switching having a greater traffic handling capacity above that point.
One of the features of DDX-1 is the signaling sequence that occurs when a call is placed which allows the caller to select either a circuit switched or packet switched call. If both calling and called terminal are of the same speed and do not require store and forward service, both terminals are served by circuit switching. If they have different speeds or require the store and forward service, they are serviced by packet switching. Signals between switching exchanges for this discrimination and other purposes of connection are themselves transmitted in packet form.
Packet transmission both between Packet Switching Exchanges (PSEs) and between computer and PSE uses the High Level Data Link Control (HDLC) procedure presently under discussion at ISO and CCITT for adoption as an international standard. Features of HDLC include transparent transmission, error control, and high transmission efficiency. The packet interleaved interface between computer and PSE uses a link number (DLN) between 1 and 255 to identify simultaneous calls. The actual header is created in the PSE.
The transmission control procedure between terminals and PSE is based upon the character-oriented HDLC procedure. Terminal messages may contain up to 1120 bits of data, i.e., 140 characters of a standard 8 bit character set.
Experimental data is reported showing various delay times as a packet switched call is set up and a packet is transmitted. It is concluded that the hybrid switching system has little overhead effect in the delay times for the packet switched network. Thus the hybrid system enables packet switching to be used for short, intermittent traffic, and circuit switching for large volume transfers, thereby providing the most efficient service for both kinds of traffic.
The European Informatics Network (EIN) project originated within Coopération Européenne dans la Domaine de la Recherche Scientifique et Technique (COST) and was formerly known as COST Project II [18]. Countries associated with the project are France, Italy, Norway, Portugal, Sweden, Switzerland, the United Kingdom and Yugoslavia, together with Euratom. The project is developing a private packet switching network with nodes at London, Paris, Zurich, Milan and Ispra (EURATOM).
Although the agreement to establish the project was signed in November 1971, it did not come into force until ratified by two thirds of the participants, which occurred in February 1973. Since then, most of the effort has been in preparing a specification for the packet switching communications subnetwork and in selecting a contractor.
The group preparing the specification for the communications subnetwork was faced with the decision about where the boundary between the user and the subnetwork should occur. The choice of boundary position determines which functions must be performed by the subnetwork, and which are the responsibility of the user. There is some variance among existing networks in this area. Eventually, it was decided to define a basic subset of facilities which would be mandatory, and another subset of facilities which were optional in use but had to be considered in the initial design. The connection between host and packet switch is to be full duplex with error detection.
The EIN subnetwork is of the datagram type, with no end-to-end control. Protocols at the host to host level are presently being defined.
As an international research project, the EIN project is not addressing the economics of packet switching. However, the project is contributing to the development of standards, the interconnection of packet switching networks, and the knowledge of packet switching in the participating countries.
The British Post Office (BPO) is currently implementing the Experimental Packet Switched Service (EPSS) as an experimental public packet switched network which will provide data to both cus~omers and the administration on the practicality and viability of packet switching for future data communications [19]. The BPO prepared the specification and contracted with Ferranti Ltd. in mid-1973 to implement the packet switch within two years.
The experimental network will comprise three nodes or Packet Switching Exchanges (PSEs), one each in London, Manchester, and Glasgow. In fact, the PSE at London will comprise three independent modules or Packet Switching Units (PSUs) to improve availability; the PSEs at Manchester and Glasgow each comprise two PSUs. The nodes will be linked by 48 Kbps lines, initially analogue but later digital. Each node will be provided with a number of ports to allow the connection to the system of two separate categories of customer terminals. One category is packet terminals, capable of constructing, transmitting, and receiving standard format packets. The other category is character terminals, only capable of operating in character mode. For such terminals, the local packet switching exchange is required to assemble characters into packets and break down packets into characters. Lines to the packet terminals will vary from 2.4 Kbps to 48 Kbps synchronous; lines for character terminals vary from 50 to 300 bps asynchronous. The BPO specification sets particularly stringent requirements for reliability of the packet switches which are being met by multiprocessor packet switches.
The packet protocol is based on the concept of a call. A call is set up by the calling terminal generating a "call originating" packet and the called terminal responding with a "first response" packet. Further packets in the call are called "subsequent" packets. Packets between a PSE and a customer packet terminal comprise a header and data. When a packet arrives at a PSE from a customer packet terminal, it is converted into a form suitable for transmission on the EPSS network by attaching a "main network addition" to the packet immediately after the data. The main network addition is removed by the destination PSE before the packet is delivered to the terminal. The length and content of the packet header and the main network addition are different for the various types of packets.
The PSE to customer packet terminal line transmission protocol allows packets to be transmitted simultaneously in both directions, but second packets cannot be transmitted until the first is acknowledged. High throughput is achieved by generating and transmitting the acknowledgment two byte times after receiving the checksum. The acknowledgment is transmitted at the fixed time in the transmitted data stream, even in the middle of a packet. Thus two main requirements of the PSE and the packet terminal are the ability to maintain synchronization between the receiver and the transmitter and the ability to measure the loop delay of the line.
A second, simplified PSE to customer packet terminal line transmission protocol waits until a packet is completely transmitted before the acknowledgment is transmitted, in the form of a packet.
The above protocols are implemented as special hardware interfaces, called a Packet Line Card (PLC) which is a microprogrammed, full duplex, synchronous line card capable of operating at transmission speeds of 2.4 Kbps to 60 Kbps.
No economic analyses supporting the development of the EPSS are known to have been published.
Research in packet switching networks at the British National Physical Laboratory (NPL) predates ARPANET, having commenced in 1966. NPL has implemented on the laboratory site a data communications network which comprises one node of a packet switching network to which a large number of computers and terminals are attached [20].
Host computers (known as User Machines) communicate with the packet switch by sending and receiving packets over a full duplex link. The packet switch offers a datagram type of service but the host-to-host protocol creates a virtual circuit. Each host is connected to the packet switch via a special interface or Network Termination Unit (NTU) which is physically adjacent to the host. The NTU is connected to the host by a pair of standard interfaces (BS4421) and the NTU can exchange special status signals with the packet switch to control the flow of packets.
Terminal devices can access the network directly by means of the Terminal Processor (TP) which appears logically to the packet switch like a User Machine.
In fact, the Terminal Processor coexists with the packet switch in the same Honeywell 516 similar to the ARPANET TIP. Terminals are connected to the Terminal Processor via a second type of Network Terminal Unit which uses either a standard CCITT interface or a BS4421 interface to the terminal. Terminal users are provided with a control panel having four illuminated push buttons for the purposes of signalling the Terminal Processor.
The packet switch and terminal processor are implemented in a 32K Honeywell 516. It has a duplex processor for reliability, providing a cold standby. Throughput is about 1 million packets per day with a peak performance of half a million bits per second. Approximately 12 computers and 85 terminals are connected. One of the host computers, also a Honeywell 516, controls a central file store comprising 60 Megabytes of disk storage.
NPL and ARPANET have been the prime catalysts in the development of most other packet switching networks. The NPL network provides a unique capability for accessing the various computers on the laboratory site from the same terminal. Experiments in interconnecting networks are being conducted, with connections to CYCLADES, EIN, and EPSS. A substantial number of professional papers on packet switching networks, as well as a book [21], have been authored by the NPL networking group.
The Reseau a Commutation par Pacquets (RCP) is being developed by the French PTT [22]. RCP is an experimental prototype network which is expected to result in the definition and implementation of a public packet switched data transmission service. The initial configuration of RCP consists of three packet switches, in Paris, Rennes, and Lyon, and three time-division multiplexers, each connecting a distant city to a packet switch.
Customer computers have access to the network through 4800 bps transmission lines over which they can have several interleaved conversations with other computers and/or terminals. Terminals can access the RCP directly through dial-up or leased lines.
Data transmission is based on the establishment of full duplex "virtual circuits" between pairs of customers. The virtual circuit is characterized by flow control at both ends and a low undetected error rate. Over such a virtual circuit, data is transmitted as a sequence of 8-bit bytes with interspersed end of message markers. The number of bytes between two end of messages is arbitrary. Messages may be split into packets or grouped consecutively within the network, but the sequence order of data bytes and the position of end of message markers is guaranteed at the destination.
The packet switches are PDP-11 models 20, 21, and 40, each with 24K words of memory. Fixed routing is used and a fixed buffer allocation is assigned for each virtual circuit, of 32 or 256 bytes per node depending on class of service.
A call request is required to set up a virtual circuit. A call request includes the address of the called customer, the address of the calling customer, a collect call flag which if set indicates the called customer is to be charged for the call, and a class of service indicator. Two classes of service are defined: low peak throughput calls with less than 120 bytes per second, and high traffic calls with peak throughput above that limit.
Although not addressed in the published paper, the author, R. Després, indicated that economic studies are being performed to determine the types of service which should be provided. However, he observed that the improved service possible with packet switching, particularly availability, is an important consideration, as well as cost. Connection of the network to commercial time sharing systems is being studied. The future of RCP may be affected by the French PTT's plans for a network called TRANSPAC.
A major objective of many of the packet switching networks being developed is to experiment with and evaluate technical aspects of packet switching. Evaluation of the economic advantages of packet switching is emphasized in only a few papers, including those on CTNE and DDX-l. This is not necessarily unreasonable, since several of the networks surveyed are currently at the implementation stage and an assessment of cost-effectiveness can be expected to receive attention later.
It is generally recognized that packet switching is more suited to short intermittent traffic while circuit switching is better for large continuous data transfers. The crossover point is put at about 5000 bits. The combination of circuit switching and packet switching when serving both kinds of traffic has been demonstrated in DDX-l.
One of the most significant selling points of packet switching is that it can provide automatic error detection, and provides increased availability of communications through alternate routing paths. Thus the packet switching network may be viewed by the user as a very reliable communications line.
Most of the papers on the networks surveyed address technical aspects of the network itself. Little attention is paid to the software required in the computers attached to the network to support user-oriented services over the network. This is reminiscent of the state of ARPANET circa 1970 when the initial IMP subnetwork had been developed but little host software was yet available. Before a network is useful to a user community, a substantial effort is required in developing protocols and implementing software to provide convenient services such as the ARPANET Telnet and File Transfer.
Considerable attention is being paid to the need to be able to interconnect packet switching networks, particularly by the European PTTs planning public data networks. Interconnection of networks appears to be more feasible with simple packet switching networks such as the datagram since functions such as flow control do not have to be performed at the interconnection point. There is a trend among the European networks to a de facto standard of 255 eight-bit bytes (2040 bits) for the maximum data length in a packet, and the use of the ISO 16-bit checksum.
The author is grateful to the following for reviewing the sections of this paper on their respective networks: L. Pouzin (CIGALE/CYCLADES), Y. Yoshida (DDX-l), D.L.A. 'Barber (EIN) , D. Wilkin (EPSS), and R.A. Scantlebury (NPL).
[1] L. G. Roberts and B. D. Wessler, "Computer Network Development to Achieve Resource Sharing, "AFIPS Conference Proceedings, SJCC Vol. 36, May 1970, pp 543-549.
[2] R. A. Scantlebury, "A Model for the Local Area of a Data Communication Network - Objectives and Hardware Organization," Proc. ACM Symposium on Data Communications, October 1969.
[3] G. D. Allery, "Data Communications and Public Networks," Proc. IFIP, August 1974, pp 117-122.
[5] P. M. Karp, "Origin, Development, and Current Status of the ARPA Network," Seventh Annual IEEE Computer Society International Conference, Digest Papers, February 1973, pp 49-52.
[6] L. G. Roberts, "Network Rationale: A five year Reevaluation," Proc. Compcon 73, February 1973, pp 3-5. Also IEEE Computer Society MAYBE
[7] L. G. Roberts, "Data by the Packet," IEEE Spectrum, February 1974, pp 46-51.
[8] D. C. Wood, "Measurement of User Traffic Characteristics on ARPANET," to be published. 4th Data Communications Symposium, October 1975.
[9] Packet Communications Inc., Application to the Federal Communications Commission, January 1973.
[10] Telenet Communications Corporation, Application for a Public Packet Switched Data Communications Network, October 1973.
[11] Computerworld, January 30, 1974, p4.
[14] L. Pouzin, "The Economics of Computer Networks The CYCLADES Case," Reseau Cyclades Report No. FIN 500, July 1974.
[16] K. Hirota et al., "A Design of Packet Switching System," Proc. ICCC 74, August 1974, pp 151-162.
[17] K. Itoh et al., "An Analysis of Traffic Handling Capacity of Packet Switched and Circuit Switched Networks," Proc. Third Data Communication Symposium, November 1973, pp 29-37.
[18] D.L.A. Barber, "Progress with the European Informatics Network," Proc. ICCC 74, August 1974, pp. 215-220.
[20] R. A. Scantlebury and P. T. Wilkinson, "The National Physical Laboratory Data Communication Network," Proc. ICCC 74, August 1974, pp 223-228.
[21] D. W. Davies and D.L.A. Barber, "Communication Networks for Computers," Wiley, 1973.