ARTICLES

NOTABLE COMPUTER NETWORKS

Computer networks are becoming more numerous and more diverse.

Collectively, they constitute a worldwide metanetwork.

JOHN S. QUARTERMAN and JOSIAH C. HOSKINS

A computer network is a set of computers using com- any one network, such as legal and social issues, and
mon protocols to communicate over connecting we conclude with a historical perspective on the
transmission media. To warrant inclusion in this ar- development of networks.
ticle, a network must provide at least mail or news
service to its users and interconnect to other net- CHARACTERISTICS OF
works that provide such services. COMPUTER NETWORKS
Among these networks are long-haul (or wide-area)
Purpose, Administration, and Funding
networks that can encompass continents. There are
In our taxonomy, there are five basic kinds of net-
also internets of smaller networks communicating
works:
with one another through the same protocols. The
internets of interest here include long-haul networks 1. Research Networks. Many of the earliest com-
as constituents, and most also include local-area net- puter networks were designed and implemented as
works (LANs). The computers connected can be research in computer networking. There are still a
small microcomputers, supercomputers, or anything number of networks that are either themselves re-
between. Multiple interconnection media can be search projects or are administered in support of
used, including coaxial cable, optical fiber, satellite other research. The ARPANET is the best-known
links, twisted pair, or telephone lines. The protocols example. Such networks are usually administered
can vary widely in speed, reliability, and general by government agencies or contractors, and sup-
functionality. The services provided can range from ported by government grants. Their users and host
the most basic mail service to distributed file sys- machines do not ordinarily pay directly for services.
tems and remote procedure call capability. A similar Because of their goals and the nature of their fund-
diversity applies to ownership, funding, administra- ing, access to these networks tends to be limited to
tion, addressing, and other characteristics. Together, researchers participating in the funded work.
these networks form a metanetwork (sometimes There are also a number of military networks,
called Worldnet) that is used daily by many commu- mostly either of the same technology or closely in-
nities of interest throughout the world. teroperable with certain research networks. In fact,
This article is primarily concerned with describing many of the research networks had as one of their
specific networks. Various characteristics are given goals the eventual development of corresponding
for each of them, where information was available. military networks. The best-known example of this
We also discuss some topics that are broader than sort of pairing is ARPANET and MILNET, which are
‘c 1986 ACM ~lOO~-078E/8G/1000-0932 7% closely allied in the ARPA Internet. Because of these

932 Cotnmurlicatiom of the ACM October 1986 Volume 29 Number 1’3

 Articles

close (and because what can be written

associations funding of such networks usually come from a single
about them is too limited to fill a separate category), company, and their users are mostly employees of
military networks are included in the research net- that company.
work category. 3. Cooperative Networks. These are networks
There are also networks that were developed to that have grown up among communities of users
provide ARPANET-like services to people and orga- with similar interests. Many of them, such as
nizations who could not obtain access to the govern- BITNET and its associated networks NETNORTH
ment-sponsored military or research networks, or and EARN, originated in an academic environment.
who did not wish to be associated with military Some originated among users of a particular ven-
work. The best-known of these (among those of in- dor’s systems (e.g., BITNET and IBM) or a particular
terest in this article) is CSNET. Users of those net- operating system (UUCP, USENET, EUnet, JUNET,
works usually pay an annual connection fee. and ACSNET among UNIX@ users) or both (FidoNet
2. Company Networks. Large corporations like among IBM PC and MS/DOS users). Many, such as
Xerox, DEC, IBM, and AT&T have implemented in- ACSNET, EUnet, JUNET, UUCP, and USENET, have
ternal networks in support of their business opera- users with a mixture of academic, corporate, re-
tions. Many of these are just LAN within particular search, and commercial interests. Often the strong-
buildings, although some are international or even est bond in a particular cooperative network (at least
intercontinental in scope. The administration and UNIX is a trademark of AT&T Bell Laboratories.

Contents
CHARACTERISTICS tiF COMPUTER NETWORKS .... 932 ROSE ........................... 950
Purpose, Administration, and Funding. ............ 932 COMPANY NETWORKS. ............ .,.. 951
Layers, Protocols, and Services ................... 934 Xerox Internet. ................... . . 951
Layerir7g Models. ............................... 934 DEC’s Easynet .................... ,. 951
Applicatiot7 Protocols. ........................... 935 IBM’s YNET. ..................... . . 952
Presenfafiorr at7d Sessio77Profocols ................. 937 AT&T.. ......................... ... 953
Trausport a77dlnter77et Protocols. .................. 937 COOPERATIVE NETWORKS ......... .... ...* 953
Network Profocols .............................. 937 BITNET ......................... .. . . . 953
Hardzoare ..................................... 937 BITNET i17the Uf7ifed States. ....... ,.. . 953
Speed and Reliability ............................ 937 NETNORTH .................... .. . . . 955
Naming, Addressing, and Routing ................. 938 EARN .......................... . . . 955
Don7air7s ...................................... 939 FidoNet .......................... . . 955
SourceRoutir7g ................................. 939 ACSNET ......................... .. . . . 956
Aftribute Lists ................................. 939 UUCP.. ......................... ... . . .. 957
Gateways.. ................................... 941 USENET ......................... . ... 958
SizeandScope .................................. 942 EUnet ........................... . . 959
Access ......................................... 942 SDN ............................. , . 961
RESEARCH NETWORKS .......................... 942 JUNET .......................... , . . 962
ARPAInternet ................................. 942 METANETWORKS ................. . . . 963
ARPANET .................................... 943 CSNET .......................... . 963
MILNETIMINET. .............................. 944 NSFnet .......................... . . 963
Other ARPA Z77ter77efNetworks. ................... 944 National Research Internet. ........ , . . 964
CSNET ........................................ 944 RARE ........................... .., .. 964
Pho77enet ..................................... 945 Pacnet ........................... .. . 964
X25NET ...................................... 945 AUSEAnet ....................... , 964
ARPANET .................................... 945 BULLETIN BOARDS AND NETWORKS .. . . 964
Cypress ....................................... 945 Commercial Networks. ............ ... 965
MFENET ...................................... 945 Conferencing Networks. ........... .., 965
SPAN ......................................... 946 SOCIAL AND LEGAL ISSUES ........ .. . , 966
MAILNET ..................................... 947 Networked Communities .......... .,* . 966
JANET ........................................ 947 Techtrical Groups. ................ * 966
EAN(X.400)Networks ........................... 948 Social Groups .................... . . 5 967
CDNnet ...................................... 948 Social Effects 1.................... I 967
European EAN Networks. ........................ 949 Legal Issues ...................... . . 967
SMARTIX/COSAC .............................. 949 HISTORY .......................... .. 967
DFN ........................................... 950 BIBLIOGRAPHIC NOTES ............ 968

October 1986 Volume 29 Number 10 Communications of the ACM 933

Ark/es

4. Commercial Networks. These networks provide

Notable Acronyms from services to outside users for profit. Some are well-
“Notable Computer Networks” known public data networks, such as TYMNET and
ARPA Advanced Research Projects Agency-the TELENET. Many are common carriers like the tele-
acronym has been changed to DARPA, phone system. Administration is always centralized,
for Defense Advanced Research Pro-
though execution may be delegated. Funding is usu-
jects Agency
CCITT International Consultative Committee on ally derived from fees charged to individual persons
Telegraph and ‘Telephony (from the or organizations for connect time or CPU time. Most
French) commercial networks do not permit free exchange of
DDN Defense Data Network-the DDN PM0 mail with other networks and are thus outside the
is the Defense Data Network Program
scope of this article.
Management Office
FTP File Transfer Protocol 5. Metanetworks. There are several projects
FTAM File Transfer and Management being directed at extending network user communi-
IMP Interface Message Processor ties by connecting existing and as yet unconstructed
IP Internet Protocol networks into metanetworks. Such networks differ
IS0 International Organization for Standard-
from internetworks both in their goals and in that
ization
NCP Network Control Protocol their constituent parts often have dissimilar proto-
PAD Packet assembler/disassembler cols even as high as the transport layer. Most meta-
PSN Packet-Switch Node networks do not yet exist: CSNET is an exception.
RFC Request for Comments-the RFCs are a
set of on-Iine documents providing in- Several networks do not fall neatly into any of
formation about the ARPA Internet these categories: COSAC, JUNET, and SDN could be
RSCS Remote Spooling and Communications classified as either research or cooperative. COSAC
Subsystem-this is the spooIing proto- is listed as a research network because it is related
col used in VNET and BITNET
SMTP Simple Mail-Transfer Protocol
to the EAN networks and DFN. SDN and JUNET are
TCP Transmission Control Protocol grouped with the cooperative networks because they
TPO-TP4 The set of transmission protocols in the are related to UUCP, USENET, and EUnet.
IS0 protocol suite
UDP Uniform Datagram Protocol Layers, Protocols, and Services
UIP User Interface Presentation
UUCP UNIX to UNIX Copy
Computer network protocols can be quite complex.
X.25 The network layer protocol in the IS0 To keep complexity manageable, protocols are de-
protocd suite signed in lay&s, building up from those near the
x.400 The IS0 mail protocol hardware to those near the users. In each layer,
XNS Xerox Network System there may be one or more protocols that peer enti-
ties on that layer can use to communicate with one
another. The interfaces between adjacent layers
are defined, and protocol designers often assume
initially) is the network or transport protocol used
that nonadjacent layers do not communicate
(e.g., RSCS for BITNET, UUCP for the UUCP mail
directly [l&l].
network, and FIDO for FidoNet).
Administration of cooperative networks is gener- Layering Models. The International Organization for
ally distributed. Some, like BITNET, have a certain Standardization (ISO) has proposed a standard refer-
amount of centralized control and organization. ence model for what they call Open Systems Inter-
Others, like UUCP, function in near-complete anar- connection [39]. This model has seven basic layers:
chy. Fees are not generally collected by a central physical, data link, network, transport, session, pre-
organization, but are paid by each node for connec- sentation, and application. The network layer is
tions to other nodes (for example, in telephone bills). often assumed to be X.25, a protocol in a series pro-
Some cooperative networks nonetheless receive mulgated by CCITT. The transport protocols, TPO
strong infusions of money from specific companies. through TP4, provide different classes of service,
For instance, BITNET has received massive contri- ranging from simple datagrams to reliable connec-
butions from IBM, AT&T spends more on UUCP and tions (TP2 is designed especially for use over X.25).
USENET than any other company (though there The higher layers are nearing design completion,
are others that also pay far more than their share), and many of them are already implemented.
and DEC and Philips have provided strong backing Much of the IS0 work is based on the work of
for EUnet. those who designed and continue to do research on

934 October 1986 Volume 29 Number 10

 Articles

the ARPANET and the ARPA Internet [8], as well as The protocol suites discussed above were designed
on related early network efforts such as CYCLADES with the assumption of dedicated links between net-
[z?]. The ARPANET originally had three basic lay- work nodes. There are other protocols, such as
ers: network, transport, and process/applications, UUCP and Fido, that were designed for use with
plus the network hardware. The ARPA Internet adds intermittent connections. These will be discussed
a fourth, internet layer, for which the IP (internet) below in sections on the networks that use them,
protocol is used. (There is also a physical layer, and along with other protocols for dedicated connections.
some descriptions distinguish a link layer plus a util-
ity layer that is similar to a combination of the IS0 Application Protocols. The types of services (applica-
presentation and session layers.) IS0 has also re- tion protocols) provided by the various networks
cently adopted an internet sublayer of the network vary widely, but they tend to fall into recognizable
layer that strongly resembles IP. The two most com- classes.
monly used transport protocols in the ARPA Internet Electronic mail is the most widespread and the
are transmission control protocols (TCPs) [53] (reli- most rudimentary. It allows a user to send a message
able connections) and uniform datagram protocols to another user on either the same or a different
(UDPs) [83] (datagrams). There is contention be- host. The message is placed in the recipient’s mailbox
tween the proponents of the IS0 Reference Model on the destination host. Mail is typically a one-to-
(ISORM) and the ARPANET Reference Model (ARM) one point-to-point communication medium, though
or ARPA Internet Reference Model [8]. The IS0 pro- it is possible to mail to more than one recipient at
tocols expect virtual circuits at the network layer, the same time. Most networks implement mail in a
whereas the TCP/IP suite makes more use of data- batched, asynchronous manner, with errors being
grams (IP is a datagram protocol). (This is an exag- reported by mail messages from a mail daemon pro-
geration for pedagogical purposes of the true situa- cess. Ordinarily, anyone can send mail to anyone
tion.) Table I shows some of the differences in with computer access.
layering as well as the layers used in a networking File transfer (sometimes known as FTP or FTAM) is
implementation in the 4.3BSD version [73] of the probably the second most common service. It allows
UNIX operating system. files to be transferred from one host machine to an-
There is a third major protocol suite called Xerox other. Since data formats vary widely among operat-
Network Services (Xerox NS) (see the “Xerox ing systems and machine types, there are usually
Internet” section), having layers that are similar several file transfer formats supported. The most
to both of the above models. There are also the generally usable one is ASCII text. The user initiat-
Coloured Book protocols, which are primarily used ing a transfer must have read access on the source
in the United Kingdom in JANET. For the purposes file and write or create access on the destination file
of this article, the four layers-network, internet, or directory.
transport, and applications-are at about the right Remote command execution (a special case of which
level of detail. is Remote Job Execution or RJE) facilities are pro-

TABLE I. Network Reference Models and Layering

Application
User programs telnet, ftp, named, time,
Process/
Presentation and libraries rlogin. or rep rwho, or talk
Applications
Session Sockets SOCK-STREAM SOCK-DGRAM
Transport Transport TCP UDP
Protocols
Network Internet IP
(Internet)
Network Ethernet
Network
Data link interfaces driver
Network lnterlan
Physical Physical
hardware controller

October 1986 Volume 29 Number 10 Communicatiozzs of the ACM 935

Articles

vided about as frequently as file transfer. They allow are some complications because different systems
a user to execute a command on a remote machine. expect different terminal types, and some networks
There are usually either strong permission checks or distinguish remote terminal access from host-to-host
a sharply limited set of commands that can be exe- remote logins. Permission checks are usually similar
cuted. to those for access by a direct terminal line.
Remote login is the simplest service in concept, Computer conferencing systems are message ex-
though many networks fail to provide it. It allows a change systems that are generally similar to elec-
user to access a system on a remote host over a tronic mail systems, but differing in that they are
network as if connected to a direct terminal. There many-to-many (or broadcast) rather than one-to-one

TABLE II. Characteristics of Some Notable Networlrs (see Table Ila for an explanation of the symbols used here)

Research networks
ARPA Internet USA 3,8 2,050 ? i T lfmno 56,000,m,9
ARPANET WA .1,1 150 ? a T Ifmo 56,000,m,9
MILNET USA ‘2,3 400 ? a 7 lfmo 56,000,m,9
MINET Europe ‘..I,4 -? ? X T lfmo 9600,m,8
CSNET USA 4,lO 170 ? i c m 1200,h,8
Phonenet USA :‘4 $26 ? d M m 1200,h,7
X25NET USA * I. )_ 38 ? X / f Ifm 9600,m,8
ARPANET (CSNET hosts) USA I’.( $,I ,25 ? I f lfmo 56,00O,m,Q
Cypress IJSA ” :+. ; 1 ,: 6 ? P T lfmo 9600,m,9
NSFNET USA ” , i,l +‘. 65 ~? i .T lfmo Tl ,m,9
MFENET USA ’ i,4. 120 ? i 0 fm 56,000,?,?
SPAN USA : I.? ‘100-b ? Px 0 fm 56,000,m,8
MAILNET USA 2,i? 28 1,800 UxiM m 1200,h.Q
JANET X IB lfmo 4800,?,?
EAN networks Europe ; 3.t2 33 ? X x m 24OO,m,Q
CDNnet Canada j n:s 1,1 32 ? X j x lmno 24OO,m,9
, ),
COSAC France i 1.2 27 ? X * x fmn 1200,d,7
DFN Germany ’ j ,I.,% , 6 ? X , x fm 9600,?,?
_’
Company networks
Xerox Internet USA : ;s,d 1 ? ,,‘K&OOO- i ‘N lfmo 56,000,h,9
Xerox RIN USA .. 1 ? 4,000 i ,. NPT lfmo 56,000,h,9
Xerox CIN USA -- 4 .?.’ 8,000 i : N lfmo 56,000,h,9
.(
DEC’s Easynet USA ., : .,4,20? iq,ooo+ SO,OQO? d iR lfmno 56,00O,m,Q
IBM’s VNET USA : ,_4,? 2,200 ? fmo 96OO,h,9

Cooperative networks
BITNET ? fmo 9600,h,8
BITNET ? fmo 9600,h,8
NETNORTH fKl0 960O,h,8
EARN fmo 9600,h,8
Asianet fm0 9600,h,8
FidoNet m 1200,d,4
ACSNET fmno 1200,h,9
UUCP mail m 1200,d,5
USENET news n 1200,-,7
EUnet lfmno 1200.h.8
SON lfmno 2400,d,8
JUNET mn 2400,d,8

936 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

TABLE Ha. Legend for Table II

Extenf: A pair of the numbers of continents and~nations reached _
Layers: Network layer (left column) and internet or transport layers $right column)
%
Protocols for the network layer
a: ARPANET-style PSN (BBN 1822) communications subnet
d: Dial-up telephone
e: An ethernet
i: An intemet over various network layers
p: Leased telephone line
x: X.25 (usually over leased telephone line)
Protocols for the internet and transport layers
A: ACSNET’s SUN-III
B: JANET’s Coloured Book
C: CSNET: TCP/IP (ARPANET); TCP/IP on X.25 (X25NET); MMDF2 (Phonenet)
0: DEC’s DECNET
F: FidoNet
N: Xerox Network Services protocol suite
P: Xerox PARC Universal Packet (PUP) protocol
T: ARPA’s TCP/IP protocol suite
U: AT&T’s UUCP
R: IBM’s RSCS
X: CClTT/lSO X.400 and related protocols
Services: I: remote login; f: file transfer; m: mail; n: news; o; other
Quality: A triple of speed, delivery, and reliability
Speed: Most typical speed of long-haul links in bits per second (bps):
300, 1,200.2,400,9,600. 19,200, 56,000, Tl microwave
Delivery: Average delivery time for mail messages:
m: minutes; h: an hour or more; d: a day or more
Reliability: On a subjective scale from 1 (lowest) to 10 (highest)

media. Usually one copy of a message is kept per tion of one of the main networks on which they are
host rather than one per user as for mail. Though used.
some conferencing systems are synchronous and im-
Network Protocols. There is not sufficient space
mediately interactive, most of the ones mentioned in
to discuss network layer protocols at any length.
this article are batched and asynchronous, since
Tables IIa and IIb indicate which ones are used by
they are distributed over wide areas. USENET news
which networks.
and bulletin-board systems are examples of this kind
of system. (See the “Bulletin Boards and Networks” Hardware. The main concern with hardware from
section.) our perspective is the speed and reliability it can
Some networks support sophisticated services confer on any particular network.
such as remote procedure call, distributed databases,
or network file systems. Most of these work best on Speed and Reliability
fast networks, however, and most long-haul net- It is difficult to find metrics of speed and reliability
works are not fast enough. Such services are not that can be applied to a range of networks as diverse
widely supported on the networks we discuss. as that covered in this article. It is even more diffi-
cult to get enough information to apply such metrics.
Presentation and Session Protocols. The IS0 model
We provide a few ad hoc measures and do not pre-
distinguishes a presentation layer and a session
tend to treat this topic in depth. For reliability, we
layer. These are of more concern to programmers
use a subjective scale from 1 (worst) to 10 (best). For
than to users of the networks, and many of the net-
speed, we use two measures: bits per second (bps)
works we discuss do not distinguish such layers
and average time of delivery of mail.
clearly. We thus have little to say about them.
Though many internets may include ethernets,
Transport and lnternet Protocols. There are a number ring networks, or other fast local-area networks, we
of protocols that are used either on more than one use bps for the most common long-haul links be-
network or on large, significant networks. We pro- tween widely separated hosts. By “most common”
vide a few details on such protocols in the descrip- we mean most likely to be used in ordinary commu-

October 1986 Volume 29 Number 10 Communications of the ACM 937

Articles

nications, not most commonly implemented. If a net- SALLY.UTEXAS.EDU a further subdomain (in this
work had many 9,600-bps links and one Tl micro- case, SALLY.UTEXAS.EDU is a host machine). The
wave link, and most traffic used the latter, we would user interface software on machines in the
list the Tl speed. UTEXAS.EDU domain may allow users to abbreviate
The average time for delivery of mail varies so SALLY.UTEXAS.EDU as SALLY. However, there
much even within networks that we list only three could be another host named SALLY.CSS.GOV, in
values: which case the abbreviation SALLY on hosts in the
domain CSS.GOV would not refer to the same host
l minutes, indicating a delivery time of less than an as in UTEXASEDU.
hour; The address 10.2.0.62 is actually a two-level ARPA
l hours, indicating a delivery time of at least an Internet address. The prefix 10 is the network num-
hour, but less than a day; ber of the ARPANET, and the rest (the local part) is
l days, indicating a delivery time of at least a day. a host number on the ARPANET. The local part can
be mapped to a network address by different meth-
We also record whether a network consists mostly of ods for different networks. In this particular case,
dial-up telephone links or of dedicated links. the network address is actually contained in the
Internet address, and there is a further hierarchy in
Naming, Addressing, and Routing the host address. The final 62 is the PSN number,
These three related terms, which are important to and the rest is the host-on-PSN number.
networking, are often confused [46, 771. The name of Routing in the Internet is also hierarchical: First a
a host, mailbox, or other resource is what a user route is found to the appropriate network through
uses to indicate the resource desired. Its address gateways (by the Gateway to Gateway Protocol
specifies the location of the resource to the network (GGP) [33] and the Exterior Gateway Protocol (EGP)
software. A route is used by the network software to [54, 76]), then a route is found to the appropriate
determine how to get to the resource. In the public host on the network (by protocols appropriate to the
switched telephone network, a name is a personal network). In the ARPANET, the latter problem re-
name, such as Jane Doe, an address is a telephone duces to finding the host’s PSN, the number of
number, and a route is a sequence of telephone lines which is encoded in the address. For an address on
and exchanges that are used to reach Jane’s number an ethernet (e.g., 128.83.138.11), finding the appro-
from the caller’s telephone. priate host is usually simpler since ethernets are
Consider hosts on the ARPA Internet. A host broadcast networks. (Ethernet (capital E) is a specific
might be named SALLY.UTEXAS.EDU and have an Xerox protocol used for LAN, whereas an ethernet
Internet address of 10.2.0.62. The address would be (small e) refers to an Ethernet-like network.)
discovered by the software on the user’s machine There are two kinds of routing: source routing,
(either by old-style static host table lookup or by where the user supplies the route to the desired
new-style domain nameserver protocols). The IP pro- resource, and system routing, where the network soft-
tocol would then use the address to route the packet ware determines a route. Most networks and inter-
to the appropriate network. The network named by nets provide system routing [26]. There are a few
the address 10.2.0.62 is network 10, the ARPANET. exceptions, most prominently UUCP. The metanet-
The ARPANET has a communications subnet of work of differing networks and internets frequently
computers called Packet-Switch Nodes (PSNs) to requires source routing to reach the appropriate net-
which hosts are attached. The PSNs then extract an work because there is as yet no universally accepted
ARPANET address (host 2 on PSN 62) from the IP network addressing convention. Source routes like
address and use it to determine a route to the desti- “alpha!beta%gamma@delta” are thus unfortunately
nation PSN and thus to the destination host. Note still common.
that names and addresses are relative to network A resource may have more than one name,
protocols. The IP address is treated as a name when address, or route. In the ARPA Internet,
the ARPANET address is extracted from it. Routing SALLY.UTEXAS.EDU might have two addresses,
is first done on the IP address and then on the 10.2.0.62 and 128.83.138.11, if it were connected to
ARPANET address. two networks. Though hosts in the Internet have
Naming, addressing, and routing can all be hierar- only one primary name, they may be known by
chical. SALLY.UTEXAS.EDU is an ARPA Internet other names on non-Internet networks. For instance,
domain name, where EDU is a top-level domain, SALLY.UTEXAS.EDU might be known as ut-sally on
UTEXASEDU a subdomain of EDU, and the UUCP network. It would be better if every host

938 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

had one name for all networks, but that is not yet administrator to run a program (pathalias) on a large
possible. Both the IP protocol and the ARPANET database of information about most of the hosts in
network protocol are datagram based, and different the network and which other hosts they connect to.
datagrams can pass through different routes to reach The result is a database of source routes from the
the same destination, even when the source is the local machine to all the machines that are in the
same. connectivity database. The local mail system uses
the source route database to convert addresses into
Domains. The ARPA Internet domain name system
routes.
is an attempt to decentralize administration of the
Problems with this scheme include the extrava-
mapping of host names to host addresses by the use
gant use of CPU time and disk space required to run
of nameservers, each of which controls part of the
pathalias and keep copies of the databases, plus the
name space [57, 59, 70, 711. This became necessary
fact that the connectivity database is published only
partly because the static host table formerly used for
once a month and is thus guaranteed to always be
that purpose had become unwieldy with the growth
out-of-date.
of the Internet and partly because most of the hosts
in the Internet are on networks local to particular Attribute Lists. Names and addresses can be either
organizations thus making it desirable to allow the absolute or relative. In the ARPA Internet, both In-
local administration to control that mapping. The ternet addresses and fully qualified domain names
domain name system also implements a hierarchical are absolute (within the Internet), but user mailbox
naming scheme and provides protocols for commu- names are relative to domain names. Most other net-
nication with the nameservers [55, 56, 58, 671. works have absolute names and addresses (again,
The British network JANET has a domain system UUCP is an exception).
similar to that of the ARPA Internet, but with the Relative names are a problem because they make
domains in the opposite order. The root is on the left mapping into addresses ambiguous. This is why
rather than on the right. The Australian network short names like SALLY are considered to be only
ACSNET also has an Internet-like domain name abbreviations for a single primary name such as
system. SALLY.UTEXAS.EDU; it is the responsibility of the
At a recent meeting, North American representa- local user interface to produce the primary name
tives of the ARPA Internet, BITNET, CSNET, and when communicating with any other host. Relative
UUCP decided to adopt the ARPA Internet domain addresses are a problem because a host may have a
naming syntax and domains as a common naming different address depending on where it is being ad-
syntax [68, 691. (The adoption is voluntary on a per- dressed from. Both relative names or relative ad-
host basis on UUCP and BITNET.) EUnet in Europe dresses leave open the possibility that two hosts
is moving in the same direction and has already might have the same address, which would make
registered several top-level national domains. JUNET proper routing impossible. Nonetheless, maintaining
in Japan already has a similar domain system. A absolute names is difficult, since absolute really
metanetwork, NSFnet, has also standardized on means relative to some standard, and there is no
TCP/IP and related protocols. Thus Internet do- universal standard. X.400 is one attempt to handle
mains may become the de facto standard, at least this problem.
in the United States. The UUCP network has not had absolute host
The IS0 X.400 mail standard also has a domain names or addresses. A single name (e.g., bilbo) may
system, which uses attributes. A resource is defined be assigned by several different companies to several
by a name and several attributes. Name conflicts can different machines. This may happen because a
be resolved by specifying sufficient attributes [39]. company was not connected to the general UUCP
There is a similar mechanism at the network level network at the time and thus was unaware of the
in X.175. conflict, or because a host was not originally ex-
The EAN networks use a simplified version of the pected to communicate with the world at large, or
X.400 system because there is as yet no registry for because the first bilbo was not listed in the UUCP
X.400 domains. map, or for other reasons.
Source Routing. Possibly the only widespread net- One method for disambiguating such conflicts is to
work without system routing is UUCP. Practically refer to each bilbo by a route from a well-known
everyone agrees that users should not have to supply neighbor (e.g., princeton!bilbo or ihnpJ!bilbo). These
source routes manually. The usual means of avoid- partial routes are a kind of attribute list in the X.400
ing this on the UUCP network is for the system sense. Of course, if someone names another host

October 1986 Volume 29 Number 10 Commurrications of the ACM 939

Articles

TABLE III. Address Formats and Gateways

ARPA Internet [email protected] u%[email protected] u%[email protected]

CSNET Phonenet [email protected] [email protected] u%h.mailnet@mit-multicsarpa:
MAILNET u%d.l@mit-multics u%d.l%relay.cs.net@mit-multics ?
JANET u%[email protected] ? 7
EAN [email protected] [email protected] [email protected]
COSAC adi/u%[email protected] aUi/u%[email protected] ?
BITNET [email protected] [email protected] [email protected]
ACSNET u%[email protected] u%[email protected] u%[email protected]
UUCP g!d.l!u g!d.l!u g!h.mailnet!u
JUNET [email protected] [email protected] u%h.mailnet@mit-multicsarpa

ARPA Internet u%d.U@cs,ucl.acuk u%d.E%[email protected] h/u%[email protected]

CSNET Phonenet u%[email protected],uk u%[email protected] h/[email protected]
MAILNET ? u%[email protected] ?
JANET [email protected] [email protected] ?
EAN [email protected] [email protected] h/[email protected]
COSAC ? ? h/u
BITNET u%[email protected] [email protected] h/[email protected]
ACSNET u%[email protected] u%[email protected] h/u%[email protected];r
UUCP g!cs.ucl.ac.uk!d.U!u g!d.E!u ?
JUNET [email protected] u%[email protected] h/[email protected]

ARPA Internet [email protected] u%[email protected] u%[email protected]

CSNET Phonenet [email protected] u%[email protected] u%[email protected]
MAILNET 7 ? ?
JANET ? ? ?
EAN [email protected] u%[email protected] u%[email protected]
COSAC ? ? ?
BITNET [email protected]~ u%[email protected] u@vnet
ACSNET u.R%[email protected] u%[email protected] u%h%[email protected]
UUCP parcvax!u.R decwrl!h.Uec.com!u g!ibm.com!u%h
JUNET [email protected] ’ u%[email protected] u%[email protected]

ARPA Internet t&[email protected] [email protected] u%h.uucp@g u%d.J%[email protected]:s.net

CSNET Phonenet u%[email protected] ;@d.A u%h.uucp@g u%dJ@utokyo-relay
MAILNET ? ; u%d.A%g@mit-multics ? u%d.J%csnet-relay@mit-muftics
JANET ? I w%[email protected] ? u%[email protected]
EAN [email protected] [email protected] [email protected] u%[email protected]
COSAC adi/u%h.bitnet@relay.~s,net ? adi/u%h.uucp adi/u%[email protected]
BITNET u@h u%d.A@g hl!h2!h!u@psuvaxl u%[email protected]
ACSNET u%[email protected]~ [email protected] u%[email protected] u%[email protected]
UUCP psuvaxl!h,bitnet!u : $eismo!munnari!d.A!u hl!h2!h!u g!d.J!u
JUNET [email protected] [email protected] [email protected] [email protected]
Notes: From UUCP to CONnet ubc-enn is a gatewag; from El&et to the European EAN networks there is one gateway per country; there is more than one
gateway between BITNET and UUCP; LJUCP,El&t, and SDN are simitariy addressee, so only one of them is listed here.
Abbreviations: u: user: h: host; g: gateway (unnamecfhere); 8: &main.
Omissions: From company networks, with commercfel netw~ks, with the ARPA Intenet.

940 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

“princeton,” or if Princeton leaves the network, a and usually have to be considered by the user when
longer or different partial route would have to be sending mail across such network boundaries. (Mail
given for that bilbo. This problem occurs with all is often the only service that can be used.) In some
attribute list schemes: Names and addresses are not cases, such gateways may not be known. In others, it
absolute. may not be possible to reveal them because of politi-
Another possible solution, now being worked on cal or economic considerations. Table III is a compi-
by a group called the UUCP Project, is to give each lation of likely mail routing syntaxes between some
UUCP host an ARPA Internet domain name, such as of the networks discussed in this article. However,
bilbo.princeton.edu. The former UUCP name would gateways are subject to change, and the nature of
still be used as a kind of network address. Routing the information makes it impossible to compile a
would be done from domain to domain, so network- table that will be accurate for very long. In particu-
wide tables would only be needed for routes to lar, addresses using a percent sign (W) to indicate
domain gateway hosts, and complete connectivity indirection through a relay host (a kind of source
information would only be kept on hosts within a routing) are a kludge that most people hope will
subdomain by those same hosts (similar methods are be temporary. A specification (RFC987 [45]) has
already used in EUnet). The UUCP network would recently been formulated for translation between
thus be integrated into the ARPA Internet domain ARPA Internet domain addresses and X.400 attribute
name system. This plan is opposed by some people addresses. Software now exists to do that translation
who actually like UUCP source routing. For an in- and also to translate between X.400 and EAN ad-
teresting discussion of related issues by a prominent dresses. When such software is in general use, per-
party on each side, see [l]. We should point out that cent sign source routing should no longer be neces-
source routing, attribute lists, and domain names are sary between those kinds of networks.
not mutually exclusive-at least not on the UUCP The user interface may even vary among systems
network. Each can be used in combination with the on the same network. The examples in Figure 1
others. (courtesy of Christian Huitema and Steve Kille) ad-
dress the same person.
Gateways. There are several related and somewhat User interface presentation (UIP) refers to the rep-
controversial terms related to machines that inter- resentation of an address to the user. The first three
connect networks. These include repeaters, bridges, examples are for networks whose internal naming
packet routers, relays, and gateways. Most of these formats use ASCII text and also are the same as the
operate below the upper layers of protocols and are UIP. The next five examples represent the same bi-
transparent to the users. Here we are concerned nary X.400 encoding, and the last two represent the
with gateways between networks with dissimilar in- same EAN address. The binary encoding of X.400
ternet layers. These usually work less well than addresses allows all networks that use it to commu-
gateways at lower layers, are often less transparent, nicate, but there is no single standard human-

[email protected] Via ARPA Internet

. . . tucl-cs!steve
. Via UUCP
[email protected] Via JANET
gb/bt/des/steve(ucl/cs) X.400, GIPSI (of INRIA) UIP
/C=GB/ADMD=BT/PRMD=DES/O=UCL/ X.400, RFC987 UIP
OU=CS/S=Kille/
(C=gb;A=bt;P=des;O=ucl;S=steve;OU=cs) X.400, another UIP
steve!ucI!cs&des%bt&gb X.400, DFN UIP
steve!ucl!cs#des&bt.gb X.400, EARN/X.400 gateway UIP
[email protected] EAN, RFC822 UIP and domain order
/C=/ADMD=/PRMD=UK/DD.=cs.ucl.ac/ EAN, X.400 encoding, RFC987 UIP
DD.=steve/

FIGURE1. Sample Addresses for Different Networks

October 1986 Volume 29 Number 10 Communications of the ACM 941

Articles

readable text UIP. Confusion results from different l Finally, a site is a place (such as a building, com-
user interface software, from differing addressing pany, or campus) where a group of network nodes
syntaxes peculiar to specific networks, from at- is located. However, the term has a more specific
tempts to represent one network’s syntax in anoth- meaning in CSNET, whereas UUCP, USENET, and
er’s, and from attempts to encapsulate one network’s EUnet users often use the word “site” to refer to a
syntax inside another’s host. To avoid confusion, the latter usage does not
The moral of all this is that there is no magic occur in this article.
formula to get mail between any two points in
Most of the networks we describe have wide geo-
Worldnet. It’s a jungle with trails that may cross and
graphical extent, but the distribution of their hosts
conflict, lead to the wrong place, or become over-
(or users) is not uniform. Many of the internets con-
grown.
sist of many local-area networks connected by a few
long-haul networks. Thus the hosts cluster on the
Size and Scope
local-area networks, which themselves tend to clus-
It is difficult to find a single metric for size that is
ter. Most networks in North America have concen-
meaningful on all networks. The traditional unit
trations of hosts in Silicon Valley near San Fran-
is number of hosts. This is useful for networks
cisco, Route 128 near Boston, and in the Toronto
like ARPANET or CSNET, where most nodes are
area because computing-related companies in North
medium-size time-sharing systems and the exact
America tend to be concentrated in these areas. Pri-
number of users on each is hard to determine. Some
marily academic networks such as MAILNET and
networks consist primarily of workstations (Xerox
CSNET are widely dispersed, with nodes mostly at
Internet) or personal computers (FidoNet) where
academic institutions. USENET and UUCP have con-
there is usually one user per host (though many
centrations in New Jersey because of AT&T.
FidoNet nodes are bulletin boards that may have
many users). Others, such as BITNET and its rela-
Access
tives, consist mostly of large IBM and Digital Equip-
Networks generally have rules (or at least guide-
ment Corporation (DE(?) mainframes that are hosts
lines) controlling access to their services. In the de-
in the ARPANET sense, but have many more users
scriptions of the networks, we list a network admin-
per host. Also, the number of users who have access
istrator to contact for information regarding access
to a network is not usually the same as the number
guidelines and more detailed information about that
who actually use the network. Thus the number of
specific network whenever possible. For some net-
active mailboxes, for instance, may be interesting,
works, we have no published references to cite. In
but is usually hard to determine.
many such cases, someone associated with the net-
The most common unit of measurement we use in
work’s administrative organization was the source.
this article is number of hosts. Where possible, we
also give number of users or such other measures as
RESEARCH NETWORKS
we can find.
It is useful to distinguish several common terms as ARPA Internet
used herein: The ARPA Internet is an internetwork of several
networks all running the TCP/IP protocol suite [51],
l A machine is a computer of any size.
connected through gateways, and sharing common
l A system is a computer system of any size; this name and address spaces [8]. The ARPANET is the
term is usually used synonymously with machine. oldest of the networks in the ARPA Internet. Both
l A node is any vertex of a graph representing a are named after the Defense Advanced Research
network, that is, any machine (or system) on a Projects Agency (DARPA), which is part of the U.S.
network. Department of Defense. DARPA (formerly known as
l A host is a network node that has resources of its ARPA) has long been a major sponsor of networking
own (such as disks, user mailboxes, or user ac- research.
counts). A host is not a node (such as an X.25 PAD Internet with a capital 1 refers to a specific inter-
or an ARPANET TAC) used only to connect across net, usually the ARPA Internet, whereas internet
the network to other nodes. Nor is a host a gate- with a small i can refer to any internetwork. (There
way. A single machine or system may serve both is also the Xerox Internet, which uses different pro-
as a host and as a gateway, however. tocols and may be older.) Other networks, such as
BITNET, UUCP, EUnet, and ACSNET, are not part of
DEC is a trademark of Digital Equipment Corporation the capital-l Internet. CSNET is a special case: Part

942 Commurlicatiom of the ACM October 1986 Volume 29 Number 10

 Articles

of it is part of the ARPA Internet, and part of it is However, there are probably more than 2000 hosts
not. and tens to hundreds of thousands of users. There
The ARPA Internet exists to facilitate sharing of are special nodes called TACs (Terminal Access
resources at participating organizations and collabo- Controllers) on both ARPANET and MILNET whose
ration among researchers, as well as to provide a only function is to allow terminals (perhaps via dial-
testbed for new developments in networking. Practi- up modems) to reach other hosts in the Internet.
cal coordination of the entire Internet is provided by Every Internet host is supposed to support a com-
the Network Information Center (NIC) at SRI Inter- mand called whois that can be used to look up direc-
national and the Network Operations Center (NOC) tory information. Failing that, it is possible to telnet
at Bolt, Beranek and Newman (BBN). to SRI-NICARPA and type WHOIS. That host is run
The two main backbone networks of the Internet, by the NIC.
ARPANET and MILNET, are funded mostly by gov- A great deal of information can be obtained by
ernment grants. The campus area networks are anonymous FTP (login anonymous, password guest)
mostly funded by local organizations. There are in from SRI-NICARPA, particularly from the files in
general no per-user or per-message charges. Services the (NETINFO) directory. The specifications for
include remote login (telnet), file transfer (FTP), most Internet protocols are on-line in documents
mail (SMTP), and numerous other smaller services called RFCs (Request for Comments) in the (RFC)
(date, time, system status, Internet directory, etc.). directory ([83]; see especially RFC980 [41]). Most of
ARPANET and MILNET hosts are all connected to a the major Internet protocols have their actual speci-
subnet of PSNs, which are then connected to each fications in Military Standards for DOD lnternet Proto-
other over 56,000-bps dedicated lines, plus a few cols [53]; most relevant RFCs and MIL-STD docu-
satellite links (e.g., to Hawaii). PSNs were formerly ments are collected and published in the 1985 DDN
known as IMPS (for interface message processors). Protocol Handbook [82]. Many RFCs related to mail
Reliability is usually very high, but may suffer dur- and domains have also been posted to the USENET
ing implementation of new capabilities (e.g., during newsgroup comp.doc.
the implementation of the domain name system).
Speed suffers during peak periods, and telnet can be ARPANET. Implementation of the ARPANET began
painful then, but mail always gets through in a rea- in 1969 [15, 521. ARPA administrators noticed that
sonable amount of time, usually in minutes. their contractors were tending to request the same
The old top-level domain (ARPA) is temporary and resources (databases, powerful CPUs, graphics facili-
will vanish soon. Many hosts are already registered ties, etc.), and decided to develop a network among
in the new domains, which are the contractors that would allow them to share such
resources. This network demonstrated the viability
COM-commercial organizations; of long-haul packet-switched computer networks. It
EDU-educational organizations; worked so well it had developed into a research
GOV-civilian government organizations; utility (run by the Defense Communications Agency,
MIL-Department of Defense; or DCA) by the end of 1983, when it was split into
NET-administrative organizations for networks MILNET, a production military network, and ARPA-
such as CSNET, UUCP, and BITNET; NET, which reverted to research. In addition to the
ORG-other organizations. original goals of networking research and resource
Several networks may be in the same domain (as at sharing, researchers almost immediately began using
large universities), and a single network may have the network for collaboration through electronic
hosts in several domains (as does the ARPANET). mail and other services.
There are also domains for countries, such as UK for Policy is set by DARPA and executed by the De-
the United Kingdom and AU for Australia. There are fense Data Network Project Management Office (see
many people outside the United States (and some the “MILNET/MINET” section). The network is
within) who claim that all of COM, EDU, GOV, etc., funded by DARPA and other government agencies.
should be under a top-level domain (U.S.). There were several versions of the early NCP (for
ARPANET and MILNET have about 150 and 400 network control protocols) and several early versions
hosts, respectively. There are also numerous net- of TCP, but since 1983 the network has used the
works at universities and private companies that are fourth version of the TCP/IP protocol suite above
part of the Internet. Since many of the hosts on such the PSN network layer (known as BBN 1822 after the
local networks are known only to their parent orga- report that describes it). Most of the links between
nizations, the real size of the Internet is hard to judge. PSNs are 56,000-bps leased lines. Response is quick

October 1986 Volume 29 Number 10 Communications of the ACM 943

Arficles

(except at peak load periods) and reliability is high. some that still run NCP. The main DOD network
The name of the network is ARPANET; the name uses switched TELETYPE messages. TCP/IP mili-
of the Internet is the ARPA Internet (not ARPANET tary networks other than MILNET and MINET in
Internet). There is also an Internet domain ARPA for DDN are DISNET (Defense Investigative Network),
hosts on the ARPANET and related networks, but SCINET (Sensitive Compartmented Information
that domain exists only to ease the transition to the Network), and WINCS (WWMCCS Intercomputer
real domains (COM, EDU, etc.), and will soon vanish. Network Communication Subsystem-WWMCCS
There are about 150 ARPANET hosts, all in the stands for World Wide Military Command and Con-
continental United States. Access to the ARPANET trol System). There are local networks in the Inter-
is officially limited to organizations doing research net at military installations such as the Ballistics
funded by federal money [17]. Most potential users Research Laboratory (BRL).
will find it more productive to contact CSNET. Canada has the military network DRENET, which
is an ARPANET-like PSN/TCP/IP network and is
MlLNET/MINET. MILNET is a long-haul military
connected to the ARPA Internet.
network that was built using the results of the
ARPANET research. It split from the ARPANET in Ofher ARPA Internef Networks. There are several
October 1983, but is still connected to the ARPANET unusual networks in the Internet. SATNET uses geo-
by gateways at the internet l.ayer. These gateways synchronous satellites to provide paths between the
were originally intended to at least be able to limit east and west coasts of the United States that are
traffic between the networks to mail only. Currently faster than the usual ARPANET or MILNET land
they pass all traffic as if the networks had not been lines, and there are satellite links to Hawaii and
divided (except for a performance penalty). None- Norway. There are packet radio networks with
theless, their PSNs form two disjoint sets, and the nodes on mobile vehicles.
two networks could easily be separated if the need Many companies, schools, and government agen-
were to arise. More recently. the European nodes on cies have local networks that are part of the Inter-
MILNET have been separated into a network called net. These include ethernets, token rings, broadband
MINET, which is also connected by gateways. Al- networks, and ARPANET-style PSN networks. Some
though MILNET eventually adopts most successful Internet networks run the TCP/IP protocol suite on
products of networking research done on the top of X.25 on public data networks. There are even
ARPANET, it does not usually participate directly in point-to-point. connections over terminal lines, Hy-
such research, since it is intended to be a stable perchannel links, dial-up links, and Tl microwave
operational network and service disruptions are kept links. These point-to-point links are usually used to
to a minimum. connect higher speed networks. The speeds of such
ARPANET, MILNET, and MINET are the main local networks may thus vary from 1,206 bps to Hy-
constituents of the Defense Data Network (DDN), perchannel speeds or higher.
which is a subset of the ARPA Internet and consists Many campus-sized organizations actually have
of networks that are directly managed by the De- several local networks. Since there is no need for
fense Data Network Program Management Office people outside of the local organization to know the
(DDN PMO), an Office of the Defense Communica- details of such internal networking arrangements,
tions Agency (DCA). Funding is by the U.S. Depart- and since there is also a limit on the number of
ment of Defense (DOD). MILNET uses the same pro- networks that the Internet core gateways can han-
tocols as ARPANET, except MILNET has not yet dle, many organizations arrange that their networks
adopted domain nameservers and still uses static appear logically as a single network to the rest of the
host tables for host name to address mapping. Internet, with subnets that are known only locally
MILNET PSNs are connected by 56,000-bps leased [W
lines like ARPANET PSNs, but MINET hosts are con- To be part of the Internet, a network must run the
nected by 9,600-bps links and reliability is lower. TCP/IP protocol suite and be connected (perhaps
The gateways between MILNET and ARPANET are indirectly through another network) to one of the
currently overloaded and form a severe bottleneck. backbone long-haul Internet networks. Most are con-
Better gateway machines are expected to be in- nected to either ARPANET or MILNET (but seldom
stalled soon, There are about 400 hosts, most in the to both).
continental United States, with some in Hawaii and
Europe (the latter on MINET). There is a classified CSNET
segment of MILNET in addition to the readily acces- The purpose of CSNET is to facilitate research and
sible part. Access is determined by the DOD. advanced development in computer science by pro-
There are other classified networks, including viding a means for increased collaboration among

944 Commur~icatior~s of the ACM October 1986 Volume 29 Number 10

 Articles

those working in the field [12, 201. The developers of CSNET CIC
CSNET noticed that electronic mail was the most [email protected]
popular service on the ARPANET. They proposed a BBN Laboratories, Inc.
network to provide electronic mail only and used it 10 Moulton Street
to connect institutions that did not have ARPANET Cambridge, MA 02238
access to those that did. CSNET is currently a logical
Phonenet. Phonenet is a dial-up star network of
network consisting of several physical networks, but
about 128 hosts. A central relay computer (CSNET-
serving a single community. All parts of CSNET are
RELAY) at CSNET CIC polls hosts on the network at
administered by a Coordination and Information
mutually agreed upon times. Most such links are
Center (CIC) at BBN in Cambridge, Massachusetts [3].
over 1,200-bps telephone lines, but 2,400-bps service
The network has been self-supporting since 1985,
has recently been installed, and some connections
though initial funding was provided by a grant from
are over public X.25 links. The software used to
the NSF starting in 1981. Annual dues are collected
manage the connections is called MMDF2, though
from member organizations with rates set according
many sites use PMDF, a Pascal subset of MMDF,
to several classifications (usually either academic or
combined with sendmail or the VMS mailer.
industrial).
The only service supported on all the parts of X25NET. X25NET uses X.25-based public data net-
CSNET is mail, transferred in ARPA Internet works to support TCP/IP links. Because the TCP/IP
RFC822 format. But CSNET has in fact become a protocol suite is used, additional services such as file
metanetwork built up of several parts that vary in transfer and remote login are provided. The use of
their additional services, lower level protocols, that protocol suite also allows many of these hosts
speed, reliability, and other qualities. Some of these to be integrated into the ARPA Internet directly, if
parts do support remote login, file transfer, and other approved by DCA. There are about 18 hosts on
services. With time, CSNET users began to realize X25NET.
that electronic mail alone was not enough. Though
old-style ARPANET syntax (e.g., user@host) is still ARPANET. About 25 hosts on the ARPANET proper
in use, CSNET is moving toward the new ARPA pay CSNET dues and are thus logically considered
Internet domain name syntax. part of CSNET as well as part of ARPANET and the
ARPA Internet. However, all hosts in the ARPA In-
The network is mostly confined to the United
ternet can be reached from any part of CSNET.
States and Canada, but has links to international
affiliates in Australia, France, Germany, Israel, Ja- Cypress. Cypress is an attempt to provide
pan, Korea, Sweden, and the United Kingdom. There ARPANET-like service to academic research depart-
are about 180 hosts in all, many of which serve as ments at a cost as inexpensive as Phonenet service
gateways to internal company networks or national [42, p. 2311. It will use TCP/IP over 9,600-bps leased
networks. Perhaps thousands of hosts on such inter- telephone lines. The nodes are called IMPlets, and
nal networks can be reached through CSNET. Mem- the first ones are small VAXes [50, p. 91. Unlike
bership is limited to organizations “engaged in Phonenet, Cypress will support file transfer, remote
computer-related research or advanced development login, and other Internet services in addition to mail.
in science or engineering.” Use of CSNET for com- Cypress will more resemble a tree network like
mercial gain is explicitly prohibited. BITNET than a star network like Phonenet, though
CSNET CIC runs a nameserver, which is a direc- there will probably be some redundant links.
tory database of CSNET users and sites. It can be
accessed by mail or by remote login (from those con- MFENET
stituents of CSNET that support remote login). The This network originated in the mid 1970s to allow
CSNET Info-Server allows retrieval of numerous access to a Cray 1 at Lawrence Livermore National
documents by mail, To get a description of the infor- Laboratories [42]. It has since grown, using several
mation available and how to get it, mail a letter to underlying network and transport protocols to sup-
[email protected] with text containing the following port access to more supercomputers. The basic pur-
lines: pose of the network is to connect physics depart-
ments doing research in nuclear fusion, specifically
REQUEST: INFO
in Magnetic Fusion Energy (MFE). It is funded and
TOPIC: HELP
administered by the U.S. Department of Energy
REQUEST: END
(DOE) and mostly managed from Lawrence Liver-
This information is also available over the ARPA more.
Internet via anonymous FTP from sh.cs.net. Mail, file transfer, remote command execution,

October 1986 Volume 29 Number 10 Communications of the ACM 945

Articles

and remote login are all supported on at least parts interfaces. Much of the original hardware, such as
of the network. There are also specialized remote the routing center computers, came from NASA.
procedure calls for interactive graphics terminals. The upper layer protocols are DECNET (see the
Some of the links use DECNET@, while others use section on DEC’s Easynet). The lower layers are pro-
special-purpose protocols developed at Livermore. vided by NASA’s Program Support Communications
The use of nonstandard protocols has led to inter- Network (PSCN) and the NASA Packet-Switch Sys-
operability problems with other networks. Therefore, tem (NPSS). PSCN is a circuit switched network,
the DOE is considering moving MFENET to the that is, a collection of leased lines and microwave
TCP/IP protocol suite and perhaps eventually to the links. NPSS consists of X.25 links, some of them over
IS0 protocols. public X.25 networks. The backbone of the network
The existing links range from 9,600-bps to is four routing centers at the Goddard Space Flight
56,000-bps leased lines to 122,000-bps satellite links. Center in Greenbelt, Maryland, the Johnson Space
Speed between any two hosts depends greatly on the Center in Houston, Texas, the Jet Propulsion Labora-
intervening links. Reliability is high. Addressing in tory in Pasadena, California, and the Marshall Space
DECNET networks is discussed in the sections on Flight Center in Huntsville, Alabama. These are con-
DEC’s Easynet and SPAN. There are about 120 hosts nected by 56,000-bps links. Each router server is the
on the network, all in the continental United States center of a star of 9,000-bps links to the other insti-
except for one in Japan. Four supercomputers are tutions on the network. Reliability is becoming high.
reachable: a Cray 1, a Cray X-MP/B, a Cray 2, and a DECNET addresses consist of 16 bits, 6 specifying
Cyber 205. There are gateways to the ARPA Internet an area and 10 specifying a node within the area.
and possibly to other networks. Access is restricted Since there are only 64 possible areas, management
to DOE-funded researchers. of area numbers is very important. Within Easynet,
DEC’s DECNET-based company network, all area
SPAN numbers are in use; thus direct gateways between
Planning for SPAN (the Space Physics Analysis Net- Easynet and other DECNETs are problematic.
work) [28] began in 1980, and operations com- There are many DECNETs other than SPAN out-
menced in 1981. SPAN was originally oriented to- side of Easynet. They cooperate in assigning area
ward researchers in Solar Terrestial and Interplane- numbers, with SPAN management providing a
tary Physics, but is now expanding to serve other forum, especially for those networks interested in
disciplines. SPAN is a multimission, correlative data joining SPAN (ESA provides a similar forum in Eu-
comparison network serving projects and facilities of rope). A major task of SPAN’s routing centers is the
the American National Aeronautics and Space Ad- assignment of nodes to areas.
ministration (NASA) in collaboration with the Euro- There are currently more than one hundred hosts
pean Space Agency (ESA). These agencies have tra- connected directly to SPAN, all of them DEC ma-
ditionally set up data collection networks to serve chines. Outside of NASA, there are many participat-
specific space missions, but SPAN is mission inde- ing universities and laboratories, such as the Los
pendent, general purpose, low cost, and easy to con- Alamos National Laboratory. There are many LANs
nect to. (However, it is sometimes used to support indirectly connected to SPAN. Because other exist-
specific missions, such as the ICE mission to the ing DECNETs want to join SPAN, the total number
Giacobini-Zinner comet [75] and the encounter with of hosts is expected to reach five hundred within a
Comet Halley [27]). It is an operational network in year. There is a transatlantic X.25 link between
that it is not intended to promote the development Marshall Space Flight Center in Huntsville, Alabama,
of network technology, but it is a research network and ESA’s Operations Centre (ESOC) in Darmstadt,
in that it provides an infrastructure for space-related West Germany. A 9,600-bps link was installed in
research. It was not created in order to access super- September from Goddard to Germany, and one to
computers, but supercomputers are becoming more Japan is expected by the end of the year.
available through it. SPAN can be reached from TELENET, and there
Guidance for the network is provided by the users are gateways to BITNET and the ARPA Internet. Ac-
through the Data Systems Users Workers Group cess is limited to researchers in appropriate areas.
(DSUWG) and project scientists [29]. Direct adminis- For more information, contact
tration is done by project managers, network man-
SPAN Project Manager
agers, and routing center managers [30]. NASA pays
Code 633
for all the links, while other participating organiza-
NASA Goddard Space Flight Center
tions pay for their own host computers and network
National Space Science Data Center
DECNET is a trademark of Digital Equipment Corporation. Greenbelt, MD 20771

946 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

MAILNET by the Network Executive (NE), based at the SERC

MAILNET originated as a joint project of the Massa- Rutherford Appleton Laboratory, and by the Joint
chusetts Institute of Technology (MIT), EDUCOM, Network Team (JNT). JANET is funded by the Com-
and 15 pioneer sites, with some initial funding from puter Board for Universities and Research Councils
The Carnegie Foundation. Unfortunately, the net- [81]; both JNT and NE are part of the Computer
work is expected to vanish by the end of 1986 due to Board secretariat. No direct charges are made for
a lack of funds. MAILNET is an inexpensive mail usage. The annual Computer Board budget for
network connecting heterogeneous computer sys- JANET and university LANs is about 3.5 million
tems at academic institutions. It is run by EDUCOM pounds. (Not all university LANs are funded by this
and is a star network around a Multics machine at money.)
MIT. Institutions with MAILNET hosts are charged Local networks connected to JANET tend to be
an installation fee ($2100) and a monthly service fee either X.25 campus switches, Cambridge Rings
($190), plus usage charges based on the number and (CR82 standard), or ethernets (IEEE 802.3). The latter
length of messages sent each month. Eighty percent two are becoming increasingly popular. The long-
of all MAILNET messages cost less than 20 cents. haul network layer is X.25 over leased lines; some of
These fees pay for a high degree of support, Most these lines are digital. Higher layers are based on the
mail transfers are done by telephone dial up from Coloured Book protocol specifications:
the central mail relay machine, though TELENET or
Blue Book: File Transfer
TYMNET can also be used. CSNET’s MMDF soft-
Pink Book: Ethernet Protocols
ware is used to coordinate the calls, and ARPANET
SMTP protocols are used for addressing and transfer- Yellow Book: Network Independent Transport
ring messages in RFC822 format. The only hardware Service
required is a modem. Speed depends on the underly- Green Book: TS29 Terminal Protocol
ing transfer mechanism, but hosts are polled at least Red Book: Job Transfer and Manipulation Protocol
twice a day. Reliability is high. Grey Book: Mail Protocol
The old-style ARPANET (e.g., user@host) syn- Orange Book: Cambridge Ring 82, Hardware and
tax is used. Gateways exist to the ARPA Internet, Protocol Specifications
BITNET, CSNET, and JANET. Monthly traffic aver- Fawn Book: Simple Screen Management Protocol
ages just over 12,000 messages from 1,800 users.
Development of these protocols started in 1979.
There are about 30 hosts in the United States, Can-
They are sometimes called the Rainbow Book Proto-
ada, and Europe. For more information, contact
cols. Most long-distance links are 64,000-bps digital
BITNET: postmaster@educom or 48,000-bps analog, and subscriber lines are mostly
ARPA Internet: 9,600 bps.
POSTMASTER%[email protected] JANET has a domain name system similar to that
CSNET: of the ARPA Internet, but the order of the domain
postmaster%educom.mailnet@mit-multicsarpa name parts is opposite, with the root on the left. The
system is centrally administered and in full use.
JANET There are gateways to the British Telecom Packet
The origins of JANET lie in the interconnection of Switchstream Service (BT/PSS) and from there to
the centrally funded large university computer cen- the International Packet-Switched Service (IPSS).
ters and research establishments in the United King- There are also gateways to ARPANET (via Univer-
dom; the first service desired was remote batch. The sity College London), EUnet (via The University of
British Post Office Experimental Packet-Switching Kent), and EARN (via the Rutherford Appleton Labo-
Service (EPSS) encouraged the development of gen- ratory). All the gateways have access controls be-
eral networking protocols in this community during cause of funding considerations. Routing in the
the 1970s. A separate physical network was evolved wide-area network is via X.25 addresses; this ad-
to replace the point-to-point lines at the same time. dress space is independent of the CCITT X.121
The early network evolved into the Science Engi- space, but conforms to its requirements. Routing be-
neering Research Council Network (SERCnet) in tween the wide-area network and attached networks
1977 and was renamed Joint Academic NETwork (or is via network level relays using extended address-
JANET) on April 1, 1984. JANET was established to ing supported by the Yellow Book protocol.
provide network links to universities and research Tabulations of the number of hosts can vary, de-
institutions in the United Kingdom and net access to pending on factors such as whether PADS are
the outside world. counted, whether those on local-area networks are
Technical and administrative support is supplied included, and whether only registered ones are

October 1986 Volume 29 Number 10 Communications of the ACM 947

Articles

counted or an attempt is made to estimate the actual year grant that will terminate in 1988 [25]. The
number of connected hosts. There are about 915 grant includes support for ongoing research and de-
registered hosts including those on local-area nets, velopment, for the initial support of CDNnet, and for
but there are probably really about 1500 connected cooperation with a commercial organization (Sydney
hosts. There are only about 20 hosts directly on Development Corporation). CDNnet is to be self-
JANET. For access, contact sufficient by 1988. Annual dues will be collected
starting in 1987 with rates set according to the type
Network Executive of organization (e.g., educational, government, non-
Rutherford Appleton Laboratory profit, commercial) and according to the number of
Chilton CDNnet hosts at an organization. Organizations pay
Didcot the telecommunications costs for connections to
Oxfordshire, OX1 1 OQX other organizations. CDNnet does not have usage
England charges except to recover the costs associated with
For the Coloured Book protocols, contact JNT at the gateways and bridges to other networks.
same address. The EAN implementation conforms to CCITT and
IS0 specifications at the session (CCITT X.225, IS0
EAN (X.400) Networks 8327), transport (TPO: CCITT X.224, IS0 8072 and
There are several networks in Europe and elsewhere 8073), and network (X.25, PSTN, DECNET, etc.)
using the EAN implementation (first developed for layers. TTXP is also used as a network layer; it is
Canada’s CDNnet) of X.400 and other IS0 protocols. based on the specifications of CSNET’s MMDF.
They have an address format with a usual user pre- Mail is the basic application service provided.
sentation form that resembles ARPA Internet do- X.400 also provides receipt notification, which is
mains, but with an internal format of X.400 attribute widely used in CDNnet and in the other EAN-based
lists [16, 451. EARN is not an EAN network, but it networks. This is implemented in EAN as follows: If
will also migrate to X.400 (and other IS0 protocols) the sender requests this service, a receipt report will
by the end of 1987. This should be interesting, since be returned to the sender when the recipient dis-
BITNET in the United States is considering moving plays the body of the message. USENET news is
to the TCP/IP protocol suite (including RFC8.22 available on at least part of the network. There is a
mail). directory service for locating users of CDNnet and
other EAN-based networks. Remote login is avail-
CDNnef. CDNnet is available to workers in the Ca- able to hosts with X.25 service. EAN implementa-
nadian research, advanced development, and educa- tions exist for 4.2BSD, VMS, System V, and will
tional communities. It is autonomous of the Cana- soon be developed for VM/CM%
dian Department of Defense. The first intermachine Most long-haul links are X.25 at 2,400 bps, though
messages were exchanged on CDNnet in 1983 [43, they vary from 1,200 bps to 9,600 bps (the range
62, 871. The network currently uses the EAN imple- offered by the Canadian PDN Datapac). There are
mentation of X.400, although other X.400 software some leased lines. Mail delivery is usually accom-
may be approved as it becomes available. EAN was plished within minutes, and reliability is high.
and is developed at the University of British Colum- The CDNET address format is [email protected],
bia. Work was begun late in 1981 and tracked the where subdomain is composed of a list of one or more
standards work already in progress by IFIP and simple names separated by dots. This is actually the
CCITT. Sydney Development has the commercial format for presentation to the users. Internally, ad-
rights to EAN. The primary purpose of CDNnet dresses are represented in binary form as X.400 Ori-
is resource sharing and collaboration among ginator/Recipient names; the exact mapping is
researchers. likely to change with the next version of X.400, in
The network is administered by CDNnet Head- particular because of work on directory services,
quarters at the University of British Columbia, and though it is possible that the same presentation for-
by representatives at each member organization. mat will continue to be used. For messages from the
CDNnet policies are set by a Steering Committee outside world, the addressing is more complicated
representing its users. EAN development work has (see Table III).
been supported by grants from the Canadian Natural CDNnet has about 65 hosts, the busiest
Sciences and Engineering Research Council (NSERC) (ean.ubc.cdn) processing about 2000 messages a day.
since November 1981. The present NSERC grant is There are gateways to CSNET, MAILNET, BITNET,
under its university-industry program; it is a three- and UUCP, and close connections to the other EAN

940 Cmnrunicntions of the ACM October1986 Volume29 Number10

 Articles

networks. ARPANET can be reached indirectly livery in minutes to hours is usual, with medium
through CSNET, and ELJnet through either UUCP or reliability.
the European EAN networks. Access is restricted to Naming, addressing, and routing are the same as
organizations engaged in research, advanced devel- for CDNnet. Methods for reaching EAN networks
opment, and education. The network may not be from non-EAN networks vary greatly depending on
used for commercial purposes. the network of origin and the locations of the sender
For access, contact and the addressee. There is a gateway at CERN that
connects EARN, UUCP, and the EAN networks. For
CDNnet Headquarters access, contact
[email protected]
Computing Centre Alf Hansen
University of British Columbia [email protected]
Vancouver, British Columbia alf-hansen%[email protected]
Canada V6T 1W5 Trond Skjesol/Alf Hansen
N-7043 Trondheim-NTH
European EAN Networks. The EAN protocols have Norway
spread so rapidly into new nations that it is interest-
ing to track their progress (see Table IV). There are SMARTIX/COSAC
actually no EAN networks as such in the United COSAC (COmmunications SAns Connections) is
Kingdom or Australia: There are merely gateways a French research network. Development work
into the national networks, much as Australia has started in 1981 at the Centre National d’fitudes des
UUCP gateways. The German network is not strictly Telecommunications (CNET). Version 3 has been
EAN-based. CDNnet, the progenitor of all the others, operational since 1984 and has some restrictions.
may not remain solely EAN as other X.400 imple- Version 5 is in development now and will be a full
mentations become available. X.400 implementation, to be operational by the end
The objective of the EAN networks is to establish of 1986. COSAC is administered by CNET, which
communication links for the European research also funds the network through CNET, INRIA
community, in cooperation with RARE (p. 000). The (Institut National pour le Recherche d’hrformatique
networks use the EAN implementation of X.400. It is et l’Automatique), CNRS (Centre National de la
expected that they will migrate to other implemen- Recherche Scientifique), and Bull.
tations in a few years due to the preliminary nature SMARTIX is intended as a generalization of the
of and lack of support for the current implementa- ideas of COSAC. It will use an implementation of
tion. Most of the sites connected inside the same X.400 by INRIA. Funding for SMARTIX will come
country are linked by 9,600-bps leased lines. Inter- from the French government and will involve
domain links consist mostly of X.25 public switched CNET, INRIA, AD1 (Agence de l’brformatique), Bull,
networks, with some using 9,600-bps lines. Mail de- and CNRS.

TABLE IV. Timetable for the Development of the European EAN Networks

CanaUa CDNnet CDN-- Apr. 1983”

Mar. 1 I l!38ab
Norway UNlNEl-l UNINETT Oct. 7, 1984
Switzerland CERN CERN Nov. 2,1984
UK UK UK Nov. 27,1984
Sweden SUNET ” SUNET Dec. 9,1984
Switzerland CHUNET CHUNET June 17,lrjSS
Germany DFN DFN Aug. 22,1985
Ireland IRL IRL Nov. 12.1985
Italy OSIRIDE I Rec. 3,1985
Spain IRIS E l&o, z3,j9as
Australia AU AU ho. 23,1985
Netherlands NL NL Mar. 17, 1988
a First intermachine message.
‘Test network established.
Table courtesy of John Dernco.

October 1986 Volume 29 Number 20 Communications of the ACM 949

 Articles

COSAC uses the CCITT X.400 protocols over X.25, oped. The network layer is X.25, which supports re-
with IS0 transport and session protocols plus FTAM mote login. Most links are 9,600 bps.
for file transfer. Local links use 64,000-bps X.25 There are about half a dozen existing hosts sup-
links, and long-distance ones use TRANSPAC, the porting mail, all in the Federal Republic of Ger-
French public data network. There is a gateway many. Immediate plans are to have about 30 hosts:
with FNET (the French UUCP network; part of 10 4.2BSD, 10 System V, and 10 VMS. Gateways ex-
EUnet). It is possible to get to the ARPA Internet, ist to EUnet, EARN, CSNET, and the EAN networks.
CSNET, and BITNET from COSAC. Between CSNET For access, contact
and COSAC, the French CSNET host france.csnet is
DFN-Verein
used. COSAC has 27 hosts in France, of which about
Pariser Strasse 44
a dozen each are Multics and UNIX machines, a
1000 Berlin 15
couple each are IBMs and DEC-2Os, and one is a
Bundesrepublik Deutschland
VMS VAX. Most are in the environs of Paris or pro-
vincial capitals, though the two DEC-20s are actually There is also a newsletter published by a consortium
in Dublin, Ireland. For access, write to of people from several German networks, giving a
comprehensive view of all networking activities in
mulcnet/[email protected] (Claude Kintzig)
Germany:
seismo!mcvax!inria!gipsy!ch (Christian Huitema)
adi/[email protected] (Jean-Luc Bernard) Neueste Netz Nachrichten
Universitat Karlsruhe, Zirkel 2
or to Claude Kintzig at
D-7500 Karlsruhe 1
Projet SMARTIX Bundesrepublik Deutschland
CNET (PAA/TIM)
38-40 Rue General Leclerc ROSE
F-92131 Issy-les-Moulineaux ROSE (Research Open Systems for Europe) is the
France principal development project of IES, the ESPRIT In-
formation Exchange System [4]. ESPRIT is the Euro-
DFN pean Strategic Program in Information Technology of
DFN (Deutsche Forshungnetz) is the national re- the European Economic Commission (EEC). Work
search network in West Germany. There was early started on the ROSE implementations in 1984, with
German networking activity at the Hahn Meitner the goal of providing an infrastructure for collabora-
Institut in Berlin, where an X.25-based network tive research and development projects within
called HMInet was developed. There was also an ESPRIT, and eventually for other projects of other
academic network between two universities there. kinds in Europe. It is also a proving ground for the
The largest research network in the country is use of the IS0 protocols in an environment of heter-
BERNET, which is still in Berlin. In 1982 there was a ogeneous machines and both wide- and local-area
move to expand BERNET to be a Northern German networks. Funding comes from the EEC and goes to
network. However, a study conducted by Stanford five industrial partners who do the work: Bull of
University recommended as an alternative a na- France, GEC and ICL of the United Kingdom, Oli-
tional network to provide ARPANET-like services. vetti of Italy, and Siemens of West Germany. Some
DFN was started to implement this idea. tasks are subcontracted.
The purpose of DFN is to develop protocols and Services eventually expected under ROSE include
implementations in the IS0 suite that can be used mail, conferencing, file transfer (including text files),
for resource sharing and collaboration among re- remote command execution, and remote login. The
searchers nationwide and communications with for- UNIX operating system has been chosen as the first
eign researchers. DFN is the German part of RARE. implementation system. Initially, existing implemen-
The West German Ministry of Research and Tech- tations of protocols already in widespread use on
nology funds DFN and has about 15-20 people work- UNIX, such as UUCP, will be used (see also the
ing on the DFN project, though all implementation “EUnet” section). Eventually, all protocols will be
work is contracted out. those of the IS0 suite. Those protocols are chosen
DFN uses X.400 for mail, plus file transfer and from those recommended by the Standards Promo-
remote job entry using protocols designed for the tion and Application Group (SPAG) [80], which is a
network but compatible with the OS1 suite. The group of a dozen European manufacturers interested
EAN implementation of X.400 is used currently, in promoting common networking standards in their
though a German implementation is being devel- products.

950 Commukxtiorzs of the ACM October 1986 Volume 29 Number 10

 Articles

Remote terminal access will be accomplished by Grapevine name: Userregistry (e.g., JLarson.PA, PA
X.3, X.23, and X.29 PADS, file transfer will be by IS0 = Palo Alto)
8571 (FTAM), mail will use the X.400 series of Clearinghouse name: Name:Domain:Org (e.g., John
CCITT recommendations, session will be imple- Doe:OSBU North:Xerox)
mented as IS0 8326 and IS0 8327, transport will
XIN communicates with the outside world via
be IS0 8072 with classes 0, 2, and 3 over X.25 and
two hosts on the ARPA Internet: Xerox.COM
class 4 over CSMA/CD networks, the internet layer
andparcvax.Xerox.COM. Xerox.COM (formerly
will be IS0 8473, and the network layer will mostly
Xerox.ARPA) is the ARPANET-Grapevine mail gate-
be X.25 and X.75.
way, and parcvax.Xerox.COM is used for telnet and
The end-to-end addressing convention to be used
FTP. Several mail gateways connect Grapevine to
in ROSE with the IS0 protocols is a three-level hier-
the CIN mail system. Example addresses (from the
archy of eight octets for the name of the remote
ARPA Internet) are
network, eight octets for the system on the LAN, and
two octets for the transport selector. This allows (to RIN/Grapevine): [email protected]
gateways between networks to be the only machines (e.g., [email protected])
that need to know about the interconnection topol-
ogy of the networks. The transport selector could and
allow the user to choose UUCP instead of the IS0 (to GIN/Clearinghouse):
session service. A prototype network is just being set [email protected]
up. (e.g., [email protected])
Certain aliases are maintained at the Xerox.COM
COMPANY NETWORKS
gateway for ease of external addressing (e.g., Post-
Xerox Internet [email protected] goes to Postmaster.pa).
Xerox was a pioneer in network research and in fact The services offered by RIN and CIN include re-
invented Ethernet [91] along with the Xerox Palo mote login, file transfer, mail, remote procedure call,
Alto Research Center (PARC) Universal Packet distributed file system, distributed computation, and
(PUP) protocol [5] and the Xerox network system many others. Both CIN and RIN support many thou-
(XNS) [92] protocol suites. An internetwork among sands of machines. Grapevine, a very large distrib-
various company sites, the Xerox Research Internet uted mail and name system for RIN, supports about
(RIN), had developed by 1976. The Corporate Inter- 4000 users around the world. CIN has its own dis-
net (CIN) split from the RIN about mid 1985. tributed mail system and about 8000 users. The XIN
The CIN and RIN are highly interconnected and is international in scope, maintaining links to sites in
together form the Xerox Internet (XIN). The CIN was Japan, England, and Canada, as well as to numerous
intended as a stable backbone network for various sites within the United States. For access, contact
corporate needs, and the RIN is primarily intended [email protected].
to serve research and development.
Administration of the networks is distributed DEC’s Easynet
among several groups within Xerox. RIN is adminis- DEC maintains an internal engineering network
tered for the most part by Xerox PARC, while CIN is called Easynet. DEC began its network endeavors as
primarily run by other divisions within the Xerox one of the pioneers of the ARPANET. The company
Corporation. Both RIN and CIN are funded by the went on to develop its own network software, called
Xerox Corporation. DECNET, and to make it available to DEC customers
Several protocols are used by XIN. The XNS proto- by 1976. Easynet was started in 1978.
col is used by both CIN and RIN. RIN also uses PUP The basic network capabilities provided by
as well as TCP/IP on some of its nets. Reliability is DECNET include intersystem file access and trans-
reported to be at least as good as that of ARPANET. fer, electronic mail, intersystem resource sharing,
Higher speed on RIN is maintained in part by interprocess communications, adaptive routing, and
56,000-bps leased lines and Tl Microwave links. remote login. These are enhanced by Easynet ser-
Naming and addressing are handled differently for vices such as ELF (a DEC Employee Locator Facility),
internal and external users. The mail system in RIN Videotext Infobases, on-line network conference dis-
is called Grapevine, and the name system in CIN is cussions (interactive bulletin boards), system moni-
called Clearinghouse. For internal names, the fol- toring, and communications for general operational
lowing examples are typical: issues.

October 1986 Volume 29 Number 10 Communications of the ACM 951

Articles

Easynet is administered by DEC Corporate Tele- files. Another internal network is CCDN, which is
communications with funding provided by DEC. used solely for remote login.
Easynet uses the DECNET protocol. Reliability is re- VNET and the RSCS protocols began in 1972 as an
ported to be at least as good as the ARPANET’s ad hoc project of some IBM employees who felt that
Easynet’s speed is maintained by lo-Mbit ethernets the available alternatives did not meet their needs.
and 56,000-64,000-bps backbone intersite links, with IBM eventually adopted the new network, and in
lower speed links to sites with lower traffic require- this sense, VNET is the UUCP of IBM. VNET is
ments. Addressing is the same as for the Internet: for IBM’s internal network, providing services such as
instance, mail, remote login, and file transfer for company em-.
user%[email protected] ployees. VNET administration is run by a group
called the VNET Project Team, which was formed in
For UUCP, the correct address mode is 1978. This team maintains and sets network guide-
. .J!decwrl!enetnode.dec.com!user lines. There is also a VNET corporate office that was
{ucbvax, decvax,
established in 1982. VNET is funded by the IBM
Easynet has more than 10,000 hosts. Assuming Corporation.
that two-thirds of DEC’s employees are users, there The RSCS protocols are used for mail transfer.
are about 60,000 network users. The network is in- These are the same as the ones BITNET uses. VNET
ternational in scope, extending throughout North links are typically 9,600-bps leased lines, though
America, the Caribbean, Europe, the Near East, they vary from 2,400-bps to Tl speeds. Addressing is
Australia, New Zealand, and the Far East. similar to the Internet style, for instance,
The gateway on the DEC end is decwrl.dec.com,
a VAX 11/750 running 4.2BSD UNIX, along with From ARPA Internet: [email protected]
RHEA, an Easynet node in the same room that is VNET to VNET: USERID at HOSTNAME
another VAX 1l/750 running VAX/VMS plus Eun- This will result in either a direct transmission of the
ice with TCP/IP. These two machines share the re- mail to the IBM employee or in the generation of a
sponsibility for gatewaying mail between Easynet message that the intended recipient needs to regis-
and the ARPA Internet, CSNET, and UUCP. ter.
For access, contact Some non-IBM mailing lists (including the
johnsson%[email protected] USENET newsgroups) are gatewayed into internal
Digital Equipment Corporation IBM conferencing systems. VNET has grown from a
Distributed Systems Architecture few hundred hosts a few years ago to about 2200
550 King Street now. The VNET PVM nodes are perhaps not a pure
LKGl-2/A19 subset of the VNET RSCS nodes, but for all practical
Littleton, MA 01460-1289 purposes, they can be considered as such. VNET has
gateways to both CSNET and BITNET that operate
IBM’s VNET within the restrictions stated below. Through these
IBM has several internal networks supporting mail, gateways it is possible to send mail to any registered
remote login, and file transfer for company opera- IBM employee. File transfer and remote login are not
tions. The main internal network is actually two dis- supported for outsiders. The CSNET gateway can
tinct networks that together form what is generally support mail to any registered IBM employee on any
called VNET. The RSCS (for Remote Spooling and internal machine. The BITNET gateway at Yorktown
Communications Subsystem) network [31] comprises Heights is not as flexible. It can support mail only to
the mail and file transfer part of VNET and cur- those hosts at Yorktown Heights and Almaden (note
rently has approximately 2200 nodes. The PVM that the IBM Research Division site on the West
(Passthru VM) network provides a remote login facil- Coast has moved from San Jose to Almaden and is
ity for VNET. There are about 1100 PVM nodes. now known as IBM Almaden Research, often re-
VNET RSCS and PVM nodes are found in North ferred to as ARC) that are in the CUNY BITNET
and South America, Africa, Middle East, Europe, (RSCS) node table. There are also other BITNET
Australia, and Asia. As mentioned, there are other gateways that are more flexible, and there is a report
internal networks; for example, VIBTS (VTAM Inte- that describes them [88].
grated Bulk Data Transfer System) is a fast network IBM maintains its own internal security to provide
consisting mostly of Tl microwave links. VIBTS is limited access to VNET. For a person within IBM to
for transferring memory images during debugging be able to send or receive mail, it is necessary to
and other activities that require fast access to huge obtain an account on a VNET node; in addition, to

952 Conlmullicationsof tile ACM October 1986 Volume 29 Number 10

 Articles

exchange mail with external networks it is neces- ACCUNET is AT&T’s commercial X.25 network.
sary to register. Most professionals within IBM are AT&T MAIL is a commercial service that is heavily
VNET users, although a relatively small number are used within AT&T Information Systems for corpo-
registered to talk to the outside world via gateways. rate internal mail.
An outsider need not be registered to send mail to
VNET, but the person receiving the mail inside IBM COOPERATIVE NETWORKS
must be registered. Many of the people in the Re-
search Division (Yorktown Heights, Almaden, and BITNET
Zurich) and the various Scientific Centers are regis- BITNET (Because It’s Time NETwork) [24] is a coop-
tered, although most others in IBM are not. For addi- erative network serving over 1300 hosts located at
tional information, contact several hundred sites (mostly universities) in 21
countries, as shown in Table V, on the next page.
[email protected] BITNET is a communications link between universi-
ties and research centers with few requirements or
AT&T restrictions other than that a site must acquire a
AT&T has some internal networks, most of which leased line to facilitate the connection to another
use internally developed transport mechanisms. BITNET node and, in the spirit of a cooperative net-
Their most widely used networks are UUCP and work, be willing to serve as a connection node for at
USENET, which are not limited only to that corpora- least one new member. (Commercial members can-
tion and which are discussed later. All internal not currently communicate with other commercial
AT&T networks support UUCP-style hl!h2!h!u members, however.) This concept of virtually
source routing syntax and thus appear to the user to unrestricted access and the absence of membership
be UUCP. Within AT&T, UUCP links are typically fees not only characterizes the cooperative essence
over 1,28@bps dial-up telephone lines or Datakit (see of BITNET, but distinguishes it from other networks
below). available for interinstitutional communication such
Among AT&T’s other networks, CORNET is an in- as CSNET. Any university or college that is able to
ternal analog phone network used by UUCP and mo- connect to BITNET possesses unlimited collaborative
dems as an alternative to Direct Distance Dialing possibilities for both academic and administrative
(DDD). Datakit is a circuit-switched digital net and is purposes for faculty, staff, and students. This policy
similar to X.25 in some ways. Most of Bell Laborato- is similar to that of UUCP and USENET, except that
ries is trunked together on Datakit. On top of the DK it is more limited to academic institutions. BITNET
transport service, people run UUCP for mail and also resembles CSNET somewhat, but is not as
dkcu for remote login. In addition to host-to-host con- centralized and has never been supported by the
nections, Datakit supports RS232 connections for ter- government.
minals, printers, and hosts. ISN is the version of There are three main constituents of the network:
Datakit supported by AT&T Information Systems. BITNET in the United States and Mexico, NET-
Bell Laboratories in Holmdel, New Jersey, uses ISN NORTH in Canada, and EARN in Europe. There is
for internal data communication. BLICN (Bell Labs also AsiaNet in Japan, and there are plans for expan-
Interlocation Computing Network) is an IBM main- sion into South America. The distinctions are purely
frame RJE network dating from the early 1970s political, and mail can be freely exchanged between
when Programmer’s Workbench (PWB) was a com- any two hosts.
mon version of the UNIX operating system. Many
UNIX machines with PWB-style RJE links use BLICN BZTNET in the United States. BITNET had its begin-
to queue mail and netnews for other UNIX ma- nings in 1981 when the first two sites, City Univer-
chines. A major USENET host uses this mechanism sity of New York (CUNY) and Yale University, were
to feed news to about 80 neighbor hosts. BLICN connected on May 5. BITNET was originally used
covers Bell Laboratories installations in New Jersey, primarily for collaboration and communications
Columbus, Ohio, and Chicago, and links most com- among systems programmers at university computa-
puter center machines. BLN (Bell Labs Network) is tion centers. Since existing IBM networking software
an NSC Hyperchannel at Indian Hill, Chicago. was used, BITNET was initially a network of IBM
AT&T Internet is a TCP/IP internet. It is not a hosts. As BITNET has grown, software to emulate
major AT&T network, though some of the best- the protocols has been developed by commercial
known machines are on it. There are many ether- vendors and members of the BITNET community.
nets connected by TCP/IP over Datakit. This inter- The basic premise for establishing BITNET was to
net may soon be connected to the ARPA Internet. provide a communications network among universi-

October 1986 Volume 29 Number 10 Communications of the ACM 953

Articles

TABLE V. BITNET Hosts as of May 1,1988

BITNET U.S.A. U.S.A.

, Mexico MEX
Totals 2
NETNORTH @tada CAN 91
Asianet : ‘d&n JPN 7
EARN Austria A 6
Belgium B 13
Israel IL 38
Switzerland Cl-l 22
Germany D 130
Denmark DK 13
Italy I 31
Spain E 8
Franbe F 39
NetherlandS NL 39
+$md~ SF 7
$Wxxe GR 2
’ ~r@nd IRL 4
I Norway N 1
Portugal P 1
Sweden. S 8
I& GB 1
Totals _.n .‘ 17 363
,,_ ._ ,.
All parts of BITNET nc
nn>: ,“?i 1306
Table courtesy of Henry Nussbacher, ’ ‘. i I)’

ties with no special requirements, restrictions, or BITNET services and implementing new ones.
fees for membership. Today BITNET is used by Through this coordinated effort, BITNIC and
scholars and administrators from a variety of differ- BITDOC are able to provide a very high level of sup-
ent disciplines. Services provided include electronic port for BITNET. BITSERVE is a help facility orga-
mail (RFC822), file transfer, and interactive mes- nized and operated by BITDOC at CUNY that offers
sages. The interactive messages allow several users a BITNET news service, a user directory, and a list of
to communicate interactively while experiencing BITNET sites and computers, and is being expanded
only moderate delays, usually less than eight seconds. to include information on conferences, software,
User, technical, and administrative support is pro- and special facilities available to BITNET members.
vided by the BITNET Network Support Center, BITSERVE is presently accessible to all BITNET sites
which is operated jointly by EDUCOM and CUNY. and is being enhanced to allow access from other
EDUCOM handles the user services and administra- networks.
tive support through the BITNET Network Informa- Considering the absence of membership fees, the
tion Center (BITNIC), the purpose of which is to cost to institutions for BITNET is small and the re-
promote the use of BITNET in higher education. Ser- strictions are minimal, compared to other networks.
vices provided include an on-line directory, paper The site must provide the connection to an adjacent
and electronic newsletters, end-user documentation, site and either acquire IBM’s VM-based RSCS from a
workshops, seminars, and conference presentations. vendor or obtain emulation software free of charge
In addition to these direct user-service functions, from BITDOC. Government support provides the
BITNIC also provides administrative support by ne- funding for the support sites. IBM has provided
gotiating for software and equipment discounts and funds for BITNIC, but that funding ends soon. How-
by archiving network procedures and policies. ever, EDUCOM and the BITNET board of directors
CUNY directs the technical support, systems main- realize that BITNIC is essential to the success and
tenance, and software development efforts via its efficiency of the network. As of now no decision has
BITNET Development and Operations Center been made on how to support BITNIC, but it is clear
(BITDOC). BITDOC maintains the Support Center that a member fee will have to be initiated for
computer and is dedicated to improving existing BITNIC to remain in existence.

954 Commu,~icario~~s of the ACM October 1986 Volume 29 Number 10

 Articles

BITNET started as a network of IBM hosts, and pean nodes must be listed at all U.S. BITNET sites
most BITNET hosts communicate using an IBM com- and vice versa. The central administrative and tech-
munication environment. BITNET uses the NJE nical services are handled by one central computer
(Network Job Entry) protocol. Most hosts use RSCS. in each country (analogous to CUNYVM for BITNET).
Today there are many non-IBM hosts using emula- EARN has an administrative branch very similar
tion software to provide the appropriate protocols for to BITNIC. IBM supports EARN and its gateway to
the DEC, VMS, UNIX, and Sperry environments. BITNET. For example, IBM will fund BITNET links to
Hosts are interconnected by leased phone lines sup- several major European EARN sites until April 1987.
porting 9,606bps data transmission. The mail and Connections to other sites are welcome, but the con-
file delivery delay ranges from minutes to hours necting host must pay for such new links.
with medium reliability. An unusual feature of The gateway between EARN and other nets is the
BITNET is that there is usually exactly one path be- same as for BITNET. For example, to send mail from
tween any two hosts. A geographic map of the net- the ARPA Internet to EARN, the address format
work shows it to be a tree network rooted at CUNY. would be
This means that nodes near the root of the tree are
user%[email protected].
more important. An example is CUNYVM. If this
node goes down, the network will still run, but will There are more than 350 hosts in 17 European coun-
essentially be split into two networks, with files and tries. Gateways to several national academic net-
mail queued at the nodes nearest CUNYVM. works in Europe are planned. EARN is currently
The addressing format from the ARPANET to expanding into Turkey, Iceland, and Portugal.
BITNET is Countrv coordinators are available for each of the
18 countries currently connected to EARN. The
useu%[email protected]
large majority of nodes have installed a mailbox
Methods for specifying addresses to the mail system named INFO. Users on other networks may use the
vary widely, depending on the host. There are cur- internet address,
rently over 800 nodes distributed among several
hundred sites. The scope is the United States plus
one host in Mexico, although there are direct links
FidoNet
to NETNORTH and EARN. Gateways exist between
FidoNet uses a telecommunications package for
BITNET and MAILNET, EDUCOM, CSNET, and
personal computers that was developed by Tom
ARPANET, and there is also restricted access to
Jennings in 1983. FidoNet is basically an extension
IBM’s VNET. For access, contact
of the Fido Bulletin Board System (BBS), which pro-
BITNET network information center (BITNIC). vides the electronic mail portion and makes the Fido
BITNET: INFO@BITNIC BBS unique in that unattended mail transfer be-
ARPA Internet: tween other “nets” and their “nodes” is not often
INFO%[email protected] offered with bulletin-board services.
FidoNet provides a network for personal-computer
NETNORTH. NETNORTH provides communica-
users in the spirit of BITNET’s utility to the aca-
tions for a number of Canadian academic and re-
demic community. There are no special require-
search sites, and was designed using the same tech-
ments or fees for membership. The software is share-
nology and several of the same basic assumptions
ware. There is no distribution cost, though users are
as BITNET. The network currently has over 96
asked to donate a small sum. The network is distrib-
nodes, and direct links exist to BITNET and EARN.
uted without sources. The main services are elec-
NETNORTH and BITNET are connected by a leased
tronic mail for personal computers and access to
line between Cornell University and the University
USENET newsgroups for personal-computer users.
of Guelph, in Ontario. Plans are also underway to
Since FidoNet is part of the Fido BBS, all of the
provide connections between NETNORTH and
services of the BBS are also available (these are cov-
other Canadian networks.
ered in the section on bulletin-board systems).
EARN. EARN, The European Academic Research The administrative node of FidoNet is located in
Network, links over 150 hosts at over 100 insti- St. Louis. The USENET/Fido gateway is adminis-
tutions in 18 countries. EARN is similiar to tered by Bob Hartman at FidoNode lOl/lOl,
NETNORTH in that it is based on the same design vaxine!sparks!m!n!user. FidoNet is funded by its
principles and philosophy as BITNET. EARN is an users, who must send the Sysop (System Operator) of
integral part of BITNET, which means that all Euro- the Fido node a cash retainer (usually $5 or $10) to

October 1986 Volume 29 Number 10 Communications of the ACM 955

 Articles

cover the phone costs when a mail message is sent. protocols. Dynamic routing was added in 1980, but
Since FidoNet operates late in the evening (after only applied to mail and file transfer. SUN-II was
11 PM local time) when phone rates are lowest, the similar, but allowed intermittent [dial-up) links
cost per message is minimal. Sysops usually provide as well as dedicated ones, plus a method of layer-
the machines (since Fido and FidoNet only run un- ing SUN-II on top of other networks (such as
der MS or PC DOS-the operating systems for IBM CSIRONET).
PCs and compatibles-the machines are IBM PCs The current version is a complete redesign and
and compatible machines) and donate their time at reimplemention done in 1983 and is called SUN-III
no cost. FidoNet is not designed to be a commercial [48]. It is layered in the traditional networking man-
venture. ner and provides a message delivery service with
FidoNet uses the Fido protocol with connections implicit (system) routing and domains in order to
made at 1,200 or 2,400 bps. A FidoNet address is support higher level protocols including file transfer,
composed of the user’s name, a net (a region or host), electronic mail, news, remote printing, simple direc-
and a node (a Fido) from the available list; for tory service, and a number of experimental services.
instance, It can transfer messages in both directions simulta-
neously over full-duplex links. It supports multicast-
user Net net-number Node node-number. ing, which is useful with USENET news and also
with mail addressed to users on multiple hosts.
The bulk of the nodes are in the United States and
The transport protocol can make use of any form
tend to be clumped in metropolitan areas such as
of virtual circuit between hosts. The links currently
St. Louis, Boston, and Chicago. However, there are a
in use include leased lines, dial-up lines, X.25, and
number of nodes in Europe, a few in Indonesia, and
CSIRONET. CSIRONET is a government research
even one in Alaska. Fido BBS is installed on over
network originally developed to connect terminal
500 hosts (IBM PCs or compatibles), and new hosts
users in remote areas to a central facility: it now
or nodes are joining on a regular basis. To access
provides virtual circuits between host machines.
FidoNet, contact the local IBM PC User Group to see
CSIRONET is slow but cheap for long distances, of
if there is a local Fido node in your area. If so, they
which there are many in Australia. There is a plan
should have the dial-up number. For more complete
to migrate the system to X.400 in the next few years.
information about FidoNet in general, contact
Most links run at 1,200 bps, and reliability is high.
Tom Jennings ACSNET has a domain naming syntax [49] similar
FidoNet: Fido’s-Board Net 125 Node 1 to that for ARPA Internet domains. The domain
UUCP: OZ.AU is registered with the Internet and can be
ihnp4!encore!vaxine!spark!l25!l!tom_jennings interpreted as a subdomain, OZ, for ACSNET, of the
country domain, AU, for Australia. There are cur-
ACSNET rently no other subdomains of AU, but there are
ACSNET (Australian Computer Science Network) is subdomains within OZ.AU, many of which are for
the main network in Australia and is based on the distributed organizations. Domains are used for rout-
Sydney UNIX Network (SUN) software developed at ing in ACSNET, so connections between machines
the University of Sydney [ZZ]. The network started determine domains more than anything else. Hosts
in 1979 and connected a machine at Sydney to an- can register in any subdomain. In practice, this
other at the University of New South Wales. It cur- means major hosts are directly in OZ, and everthing
rently spans the continent and is closely connected else is in subdomains.
to networks elsewhere [47]. The purpose of the net- There are several UUCP gateways to North Amer-
work is to support mail traffic and file transfer ica and Europe, all from Melbourne and all using
among research, academic, and industry users. The X.25. There is a CSNET link to the United States.
underlying transport protocols are also used to sup- There are EAN links to Canada, the United King-
port the USENET news network in Australia. There dom, Germany, Norway, and Switzerland. In addi-
is no central administration, though this may change tion, there are several 1,299-bps dial-up links to
in the future. At present the original developers and North America (New Hampshire, New Jersey, and
the international gateway operator act as coordina- California). There are about 300 hosts throughout
tors. There is no government funding: Each host the Australian continent on ACSNET.
pays for its own links. For general information, contact
The original protocols were called SUN-I and sup- [email protected]
ported remote login, file transfer, and multiplexed seismo!munnari!postmaster

956 Communicatiom of the ACM October 1986 Volume 29 Number 10

 Articles

For connection requests, contact generated as responses to USENET news. The same
underlying UUCP transport mechanisms are also
[email protected] used to support much of USENET.
seismo!munnari!basser.oz!acsnet-request The UUCP mail network has many problems with
or the ACSNET Coordinator at routing (it is one of the few major networks that uses
source routing) and with its scale. Nonetheless, it is
Department of Computer Science extremely popular and still growing rapidly. This is
University of Sydney attributable to three circumstances: ease of connec-
New South Wales, 2006 tion, low cost, and its close relationship with the
Australia USENET news network.
Mailing lists similar to those long current on the
UUCP ARPANET have recently increased in popularity on
The name “UUCP,” for UNIX to UNIX Copy, origi- the UUCP mail network. These permit a feature that
nally applied to a transport service used over dial USENET newsgroups cannot readily supply: a limi-
ups between adjacent systems [63]. File transfer and tation of access on a per-person basis rather than on
remote command execution were the original intent a per-host basis. Also, for low-traffic discussions
and main use of UUCP. There was an assumption mailing lists are more economical, since traffic can
that any pair of communicating machines had direct be directed to individuals according to their specific
dial-up links, that is, that no relaying was done interests.
through intermediate machines. By the end of 1978, There is no central administration. To connect to
there were 82 hosts within Bell Laboratories con- the network, one need only find one machine that
nected by UUCP. Though remote command execu- will agree to be a neighbor. For people at other hosts
tion and file transfer were heavily used, there is no to be able to find your host, however, it is good to be
mention of mail in the standard reference [64]. registered in the UUCP map, which is kept by the
There was another similar network of “operational” group of volunteers known as the UUCP Project
hosts with UUCP links that were apparently outside [84]. The map is posted monthly in the USENET
Bell Laboratories, but still within the Bell System. newsgroup comp.mail.maps. There is a directory of
The two networks intersected at one Bell Laborato- personal addresses on the UUCP network [44], al-
ries machine. though this is a commercial venture unrelated to the
Both of these early networks differed from the UUCP Project.
current UUCP network in assuming direct connec- Each host pays for its own links; some hosts en-
tions between communicating hosts and in not hav- courage others to connect to them in order to
ing mail service. The UUCP mail network proper shorten mail delivery paths.
developed from the early networks and spread as the There is no clear distinction between transport
UUCP programs were distributed as part of the and network layers in UUCP, and there is nothing
UNIX system. resembling an Internet Protocol. The details of the
Remote command execution can be made to work transport protocol are undocumented (apparently
over successive links by arranging for each job in not actually proprietary to AT&T, contrary to rumor,
the chain to submit the next one. There are several though the source code that implements the protocol
programs that do this: Unfortunately, they are all and is distributed with UNIX is AT&T’s trade secret).
incompatible. There is no facility at the transport Mail is transferred by submitting a mail command
level for routing beyond adjacent systems or for er- over a direct connection by the UUCP remote com-
ror acknowledgment. All routing and end-to-end re- mand execution mechanism [36]. The arguments of
liability support is done explicitly by application that mail command indicate whether the mail is to
protocols implemented using the remote command be delivered locally on that system or resubmitted to
execution facility. There has never been any remote another system. In the early days, it was necessary
login facility associated with UUCP, though the CM to guess the route to a given host and hope. The only
and tip programs are sometimes used over the same method of acknowledgment was to ask the addressee
telephone links. to reply. Now there is a program (pathalias) that can
The UUCP mail network connects a very diverse compute reasonable routes from the UUCP map, and
set of machines and users. Most of the host ma- there is software that can automatically look up
chines run the UNIX operating system [73, 741. Mail those routes for users.
is the only service provided throughout the network. The UUCP mail network is currently supported in
In addition to the usual uses of mail, much traffic is North America mostly by dial-up telephone links. In

October 1986 Volume 29 Number 10 Communications of the ACM 957

Articles

Europe there is a closely associated network called USENET

EUnet, and in Japan there is JUNET, both of which USENET began in 1980 as a medium of communi-
will be discussed later. cation between users of two machines, one at the
The most common dial-up link speed on the University of North Carolina, the other at Duke Uni-
UUCP mail network is 1,200 bps, though there are versity. It has since grown exponentially to its cur-
still a few 300-bps links, and 2,400 bps is becoming rent size of more than 2000 machines. In the process.
more popular. When systems are very close, they are the software has been rewritten several times, and
sometimes linked by dedicated lines, often running the transport mechanisms now used to support it
at 9,600 bps. Some UUCP links are run over local- include not only the original UUCP links, but also
area networks such as ethernets, sometimes on top X.25, ACSNET, and others.
of TCP/IP (though more appropriate protocols than USENET combines the idea of mailing lists as
UUCP are usually used over such transport media, long used on the ARPANET with bulletin-board ser-
when UUCP is used its usual point-to-point error vice such as has existed for many years on TOPS-80
correction code is bypassed to take advantage of the and other systems, adding a freedom of subject mat-
reliability of the underlying network and to improve ter that could never exist on the ARPANET, and
bandwidth). Some such links even exist on long- reaching a more varied constituency. While chaotic
haul packet-switched networks. and inane ramblings abound, the network is quite
The widespread use of more sophisticated mail re- popular.
lay programs (such as sendmail and MMDF) has in- The USENET news network is a distributed
creased reliability. Still, there are many hosts with computer conferencing system [23] bearing some
none of these new facilities, and the sheer size of the similarities to commercial conferencing systems
network makes it unwieldy. like CompuServe, though USENET is much more
distributed. Users pursue both technical and social
The UUCP mail network has traditionally
ends on USENET. Exchanges are submitted to news-
used source routing with a syntax like
groups on various topics, ranging from gardening to
hosta!hostb!hostc!host!user. The UUCP map and
astronomy.
pathalias have made this bearable, but it is still a
The name “USENET” comes from The USENIX
nuisance. An effort is underway to alleviate the
Association, The Professional and Technical UNIX
routing problems by implementing naming in the
User’s Group. The name UNIX is a pun on Multics
style of ARPA Internet domains. This might also al-
[65], which is the name of a major predecessor oper-
low integration of the UUCP name space into the
ating system. (The pun indicates that, in areas where
ARPA Internet domain name space. In fact there is
Multics tries to do many things, UNIX tries to do one
now an ATT.COM domain in which most hosts are
thing well.) USENET has no central administration,
only on UUCP or CSNET. Most UUCP hosts are not
though there are newsgroups to which introductory
yet in any Internet domain, however. This domain
and other information about the network is posted
effort is also handled by the UUCP Project and ap-
monthly. USENET is currently defined as the set of
pears to be proceeding at a methodical but persistent
hosts receiving the newsgroup newsannounce (but
pace [35, 371.
see the EUnet and JUNET sections). There are about
The hardware used in the UUCP mail network a dozen hosts that constitute the backbone of the
ranges from small personal computers through network, keeping transit times low by doing fre-
workstations to minicomputers, mainframes, and quent transfers among themselves and with other
supercomputers. The network extends throughout hosts that they feed. Since these hosts bear much of
most of North America and parts of Asia (Korea and the burden of the network, their administrators tend
Israel). Including hosts on the related networks to take a strong interest in the state of the network.
JUNET (in Japan) and EUnet (in Europe), there are at Most newsgroups can be posted to by anyone on the
least 7,000 hosts on the network; possibly 10,000 or network. For others, it is necessary to mail a submis-
more. (EUnet and JUNET hosts are listed in the sion to a moderator, who decides whether to post it.
UUCP maps.) The UUCP Project addresses are Most moderators just filter out redundant articles,
[email protected] though some make decisions on other grounds.
cbatt!uucp-query These newsgroup moderators form another group in-
[email protected] terested in the state of the network. Newsgroups are
created or deleted according to decisions made after
Much information about UUCP is published in discussion in the newsgroup news.groups.
USENET newsgroups. Each host pays its own telephone bills. The back-

959 Comn~ur~ications oj t/w ACM October 1986 Volume 29 Number 10

 Articles

bone hosts have higher bills than most other hosts usually have the same name on both networks, mail
due to their long-distance links among themselves. addresses have no meaning on USENET: Mail re-
The unit of communication is the news article. Each lated to USENET articles is usually sent via UUCP
article is sent by a flooding routing algorithm to all mail; it cannot be sent over USENET, by definition.
nodes on the network [34]. The transport layer is Though the two networks have always been closely
UUCP for most links, although many others are related, there are many more hosts on UUCP than
used, including ethernets, berknets, and long-haul on USENET. In Australia the two networks do not
packet-switched networks; sometimes UUCP is run even intersect except at one host.
on top of the others, and sometimes UUCP is not There are different distributions of newsgroups on
used at all. USENET. Some go everywhere, whereas others are
The many problems with USENET (e.g., reader limited to a particular continent, nation, state or
overload, old software, slow propagation speed, and province, city, organization, or even machine,
high and unevenly carried costs of transmission) though the more local distributions are not really
have raised the possibility of using the experience part of USENET proper. The European network
gained in USENET to design a new network to re- EUnet carries some USENET newsgroups and has
place it. The new network might also involve at another set of its own [79]. JUNET in Japan is similar
least a partial replacement for the UUCP mail net- to EUnet in this regard.
work. There are about 2000 USENET hosts in the United
One unusual mechanism that has been proposed States, Canada, Australia, and probably in other
to support the new network is Stargate [89, 901. countries. The hosts on EUnet, SDN, and JUNhT
Commercial television broadcasting techniques communicate with USENET hosts: The total number
leave unused bandwidth in the vertical blanking in- of news hosts including ones on those three net-
terval between picture frames. Some broadcasters works is probably at least 2500. The UUCP map in-
are currently using this part of the signal to transmit cludes USENET map information as annotations. A
Teletext services. Since many cable-television chan- list of legitimate netwide hewsgroups is posted to
nels are distributed via geosynchronous satellites, a several newsgroups monthly. Volunteers keep statis-
single input to a satellite uplink facility can reach all tics on the use of the various newsgroups (all 250
of North America on an appropriate satellite and of them) and on frequency of posting by persons
channel. A satellite uplink company interested in and hosts. These are posted to news.newslists
allowing USENET-like articles to be broadcast by once a month, as is the list of newsgroups. Impor-
satellite on a well-known cable-television channel tant announcements are posted to two mode-
has been found. Prototypes of hardware and soft- rated newsgroups, news.announce and
ware to encode the articles and other hardware to news.announce.newusers, which are intended to
decode them from a cable-television signal have reach all users (the current moderator is Mark
been built and tested in the field for more than a Horton, cbosgd!mark). An address for information
year. A new, reasonably priced model of the decod- on the network is seismo!usenet-request.
ing box may be available soon.
This facility would allow most compatible systems EUnet
within the footprint (area of coverage) of the satellite The European UNIX network (EUnet) started at the
and with access to the appropriate cable-television April 1982 European UNIX systems Users’ Group
channel to obtain decoding equipment and hook into (EUUG) meeting in Paris, and originally connected
the new network at a very reasonable cost. Articles machines in the Netherlands, Denmark, Sweden,
wbuld be submitted for transmission by UUCP links and the United Kingdom. It began as an extension or
to the satellite uplink facility. Most of the technical application of the software and protocols used in
problems of Stargate seem to have been solved. USENET and UUCP in North America, and most
More than 90 percent of all USENET articles reach hosts, as on those networks, run UNIX. There have
90 percent of all hosts on the network within three always been and still are some marked differences,
days. Though there have been some famous bugs however. Mail and news are much more closely tied
that caused loss of articles, that particular problem together in EUnet: The backbone hosts and adminis-
has become rare. trators are the same for both, and a single name is
Every USENET host has a name. That host name used for the combined mail and news network. The
and the name of the poster are used to identify the administration of the network is much more orga-
source of an article. Though those hosts that are on nized than for UUCP and USENET in North Amer-
both the UUCP mail and USENET news networks ica, and there has always been a much stronger rela-

October 1986 Volume 29 Number 10 Communications of the ACM 959

Articles

tionship between EUnet and EUUG than there has initial connection charge, but the per-packet charge
ever been between USENET or UUCP and either accounts for more than 80 percent of the costs. Rates
USENIX or /usr/group (the two organizations in for X.25 connections are lower in Europe than in
North America1 most similar to EUUG). Many of the North America, and usually lower than equivalent
“soapbox” discussion groups are not carried on North American telephone rates for similar dis-
EUnet (due to their high costs in transatlantic and tances and connection times. But, despite uniformity
European traffic), and there are many newsgroups in charging units, X.25 tariffs vary widely within
that are only distributed within Europe. EUnet also Europe. Mail originating outside of Europe in some
differs from USENET, UUCP, and ACSNET in almost cases must be paid for by a EUnet backbone host,
every other regard, especially funding, as will be particularly when gatewaying to a national non-
seen. Yet the purpose of the network, to provide its charging network. This makes bulk mailings rather
users with modern communication facilities, partic- annoying to the gateway administrators.
ularly electronic mail and news, which are capable English is the lingua franca of the network, though
of reaching the users of as many networks as possi- many other European languages are also used. The
ble, is in line with the other networks. basic application protocols are news and mail as in
There is one EUnet backbone host in each Euro- USENET and UUCP, plus remote login where X.25
pean member country. Each such backbone host or- links are used. EUnet originally used UUCP over
ganizes communications within its country, often by dial-up telephone links like those of USENET and
maintaining direct connections to all other hosts in UUCP in North America. This arrangement did not
the country. The backbone hosts also communicate last more than about six months, as it quickly be-
among themselves across international boundaries. came evident that X.25 links were more practical
The whole set of backbone hosts is the backbone of (faster and cheaper) in Europe for long-distance
the network. There is a central host to which all links. Most EUnet links between backbones and out-
backbone hosts have connections and that carries all side Europe are now UUCP (without much of the
the intercontinental news and most of the mail usual error checking) over X.25. Ordinary UUCP
traffic; this host has always been mcvax in Amster- telephone dial-up links are still the most common
dam. for local links and to leaf nodes.
The administrators of the backbone hosts hold EUnet has recently concluded an arrangement to
meetings (usually at EUUG meetings) where they redistribute the ACSNET software within Europe,
determine concerted strategies and tactics. Currently and it appears likely that many links between back-
they are moving toward implementing ARPA Inter- bone hosts will soon use ACSNET rather than UUCP
net RFC886 domains in EUnet and expect to have because of such obvious advantages as multiplexed
completed doing so by mid 1987. The top-level do- and full duplex connections.
mains being chosen are the ISO- two-letter The SLIP software that allows TCP/IP over serial
country codes, following RFC920. When a single le- lines may be used over some interbackbone links in
gal or political entity is needed to speak for the net- place of X.25 because of its advantage in speed
work, EUUG does so. There is some financial sup- (4,800 bps is the practical limit for X.25 links), espe-
port from EUUG, but most funds are provided by the cially over some of the faster digital telephone links
owners of the individual hosts. For example, the cost now becoming available (up to 50,000 bytes at about
of the news connections with North America is $0.25 a minute). EUnet will probably eventually
shared proportionately among the EUnet backbone adopt X.400 for mail, using whatever transport
hosts according to the number of news hosts each mechanisms, such as TP4 or other IS0 transport pro-
feeds. They in turn share these costs equally among tocols, are appropriate at that time.
all the hosts in their country. Thus no host bears a Most dial-up telephone links are still 1,200 bps,
disproportionate burden. but 2,400 bps is becoming more common. X.25 links
Mail is charged to the originating host on a are mostly effective about 4,800 bps, though the
message-by-message and link-by-link basis. Except nominal rate most commonly used is 9,600 bps.
this is not possible for intercontinental links and Ninety percent of all mail and news traffic arrives
links to nonchargeable networks. The originating within one day. Reliability is quite high.
host usually brings these charges to the attention of EUnet has already almost completely eliminated
the senders of the mail in some manner. the old-style UUCP hosta!hostb!host!user syntax in
The European Public Data Networks charge per favor of user@host. Routing is managed by the back-
segment (maximum 64 bytes] for use of X.25. There bone hosts, each of which knows the organization
is also a negligible connection time charge and an within its own country and which hosts are in

960 Communications of the ACM October 1986 Volume 29 Number 10

which country. Routing information is automatically same place. Elsewhere, direct inquiries to EUUG:
exchanged daily between the backbone hosts and
between Europe and the United States and Japan. If EUUG secretariat
mail is sent from a nonbackbone host that lacks a Owles Hall
direct link to the destination host, it is forwarded to Buntingford
the national backbone host, which relays it. Herts SG9 9PL
ARPA Internet RFC886 domain naming syntax is England
currently being implemented on EUnet. Each coun- mcvax!euug
try will register as a top-level country domain with
the Internet (e.g., NL for the Netherlands). There SDN
will also be subdomains from the beginning. Thus The System Development Network (SDN) [lo, 111
mcvax might become cwi.ac.nl, and an address for a was started in 1982 in the Republic of Korea with
user of that host there might be [email protected]. one node at Seoul National University and another
This simplifies routing further, since each backbone at the Korea Institute of Electronics Technology. The
host need then only know the hosts within its own major development issues in SDN during the initial
country domain and a path to the backbone host for period from 1982 to 1984 were setting up the envi-
each other country domain: There is no need to ronment for computer communications and adding
know anything about the internal structure of other new nodes. SDN is now the backbone that intercon-
country domains. In fact, routing is delegated fur- nects local-area networks at major sites. Most of the
ther, since the internal structure of a given subdo- nodes are connected via the TCP/IP protocol suite
main may be known by a host other than the na- through leased lines, and advanced research is being
tional backbone host, which need then only know carried out in network softwares, international
the appropriate nameserver host for the subdomain. standards, and distributed systems. The intent of the
EUnet has about 999 hosts and extends throughout network is to provide a facility for computer com-
Western Europe, as shown in Table VI. There are munications and resource sharing, and a test envi-
connections to EARN, JANET, DFN, and other net- ronment for research and development communities
works within Europe, plus intercontinental connec- in Korea.
tions to Japan and Korea, to CSNET, UUCP, and Technical and administrative support has been
USENET in North America, and to ACSNET and provided by Korea Advanced Institute of Science
USENET in Australia. New hosts must register with and Technology (KAIST) since 1983. The Network
their national backbone host administrator, and in- Management Center, located at KAIST, handles in-
quiries within Europe should be addressed to the formation dissemination and also maintains interna-
tional and domestic contacts for administrative mat-
ters. Managerial decisions are made by the Oversee-
ing Committee, which consists of representatives
TABLE VI. EUnet Connected Hosts as of Julv 1,1986 from each site. The Electronics and Telecommunica-
tions Research Institute and the Data Communica-
Austria 7 tions Company of Korea, along with KAIST, are the
Belgium 17
Denmark 36
major participants in the management and develop-
Finland 47 ment of SDN.
France 94 Each site is charged the cost for its connections,
Germany 90 whether through leased lines, X.25 PDN, or dial ups
Greece
Ireland
for domestic communications. International commu-
6
Italy 28 nications costs are charged proportionally. Expenses
Netherlands 129 for protocol development and management and in-
Norway 7 ternational communication are covered by national
Sweden 108 research grants, public corporations, and internal
Switzerland 50 funding from several institutes.
U.K. 276
Total 896 The standard protocol architecture in SDN is
Total throughput for June 1986: 550 Mbytes. Total throughput for based on the U.S. DOD IP, and UUCP is also sup-
April 1986 of one backbone host: 500 Mbytes; average throughput ported over all hosts. Some UUCP links run on top
per host: IO Mbytes per month. Split news/mail is about 50/50. Costs
per host: average U.S. 5150 for data transport. of TCP/IP and X.25. Virtual terminal, file transfer,
Statistics courtesy of Teus Hagen and Piet Beertema mail (in both Korean and English), remote command
execution, net news, and name services are sup-

October 1986 Volume 29 Number 10 Communications of the ACM 961

 Articles

ported. The SDN research community is currently Seoul 131

working on migration to IS0 protocols starting from Republic of Korea
the network and transport layer. The low-level net- UUCP: seismo!kaist!nmc
work is built largely out of leased phone lines and ARPA Internet: nmc%[email protected]
the domestic X.25 network, in addition to LANs. The
international c:onnections are based on X.25 and JUNET
X.28/X.29 PAD. Most links run at 2,400 bps, and The earliest large computer network in Japan ap-
some links run at 9,600 bps. Reliability is high. pears to have been N-l [40], which is an interuni-
The top domain name, SDN, has been used for versity network that started late in 1970. It currently
years in SDN. !<econd-level domains are usually connects about 20 national universities using DDX-P
hosts. Routing is decided at the issuing host. In addi- (Digital Data exchange, Packet-switching network),
tion to the domain style name, UUCP-like host!user which is the Japanese Public Data Network. (There
names can be used wiih a pafhalias database. is also a circuit switching network, like Canada’s
Internet dom.ain naming as proposed in RFC882 Infoswitch.) N-l provides only remote login and no
and RFC883 is being implemented. The naming mail service.
structure consolidated as of June 1986 is in confor- JUNET originally linked three universities starting
mance to RFC920. The top-level domain is KR for in October 1984. It was connected to Europe in Janu-
Korea, with the following second-level domains ary 1985 by a link between kddlab and mcvax; con-
below it: nections to other continents have since been added.
It is currently the major nationwide noncommercial
RES for research community computer network in Japan (DDX-P might be larger).
EDU for educational institutes JUNET incorporates both news (like USENET) and
COM for companies mail (like UUCP) in a single network organization
GOV for governmental organizations (like EUnet). Its purpose is to promote information
ORG for genera.1 organization exchange among Japanese researchers and with re-
<network-names> for other nationwide networks searchers outside Japan [61].
Third-level domains are the organization names. The network is organized by a group of students at
Fourth-level domains usually apply to hosts. Tokyo Institute of Technology, whose leader is Jun
UUCP gateways to North America were set up in Murai. The administrators of the major (backbone)
1983 over X.25 and dial-up lines. A CSNET link us- hosts on the network also help administer the net-
ing PMDF over X.25 was installed in 1984. The work and hold monthly meetings for that purpose.
CSNET connection will soon be replaced by TCP/IP There is evidently a tradition in Japan of employ-
over the X.25 link. ees asking their employers for permission before
Within the Republic of Korea, about 100 com- publishing anything publicly; this might explain the
puters at 22 different organizations were connected small number of Japanese news postings seen out-
to SDN as of June 1986. By the end of 1987, ZOO-300 side Japan. Also, there is a distribution (fj.all) of
computers at 30 different organizations will be con- newsgroups that can be seen only inside Japan.
nected. SDN administrators are now working on For instance, fj.kanji (for kanji handling) and
Pacnet [9], a cooperative network over UUCP con- fj.micro.mac are very active. Each host’s connection
nections that w:ill interconnect universities and com- costs are paid by its institution.
panies in the Pacific region. SDN currently has con- UUCP is the common protocol, with X.25, tele-
nections to Japan (kddlab) and Indonesia (indovax) phone dial ups, and ethernets below. Some UUCP
links are carried over TCP/IP. The IS0 protocols are
directly, and to Australia (munnari) and Singapore
(tataelxsi) through seismo. not yet used, and there appear to be no immediate
SDN liaison is Kilnam Chon: plans to use them, although implementations exist
within several corporations. Mail (per RFC822) and
UUCP: seismo!kaist!chon news (as in USENET) are supported. There are three
ARPA Internet: chon%[email protected] standard human-language notations supported:’

The mailing address is English in 8-bit ASCII:

We had a very good time at Mark’s home last
SDN Management Center night.
Department of Computer Science
Korea Advanced. Institute of Science and Technology ‘Thanks to Hide Takuda and fun Murai for these examples; JLSis Japan
Industrial Standard. and C6228 defines Kanji characters of 2 bytes each of
P.O. Box 150 Chongryang 7bits.

962 Communications of tlw ACM October 1986 Volume 29 Number 10

 Articles

Japanese in roman characters in 8-bit ASCII: academic community with the kind of networking
Watashi tachiwa sakuban Mark no iede tanoshii resources that CSNET now provides to computer sci-
hitotoki wo sugoshita. ence researchers [3]. It extends the ideas of resource
Japanese in a 16-bit JIS C6228 encoding of Kanji sharing, which motivated the ARPANET, and of col-
characters: laboration among researchers, which grew out of the
development of the ARPANET [19]. NSFnet will also
concentrate more on customer support and informa-
tion than has been customary in the ARPA Internet.
The mail and news interface software has been mod-
The NSFnet project is administered by the OASC
ified to support Kanji characters, which also require
in coordination with the administrations of existing
special display hardware.
networks. The NSF will provide initial funding, al-
Most JUNET links are 2,400 bps, and reliability is
though there is potential for the network to even-
high. There is a domain system, and its top-level
tually become self-supporting.
domain is JUNET. Second-level domains are usually
Diverse network protocols will be used to support
named for organizations such as universities, and
the TCP/IP protocol suite; the network may even-
third-level domains are usually hosts. This scheme
tually migrate to the IS0 protocols. Initially, NSFnet
is similar to RFC733 [14], the predecessor of the cur-
will use the existing long-haul (ARPANET) and cam-
rent ARPA Internet domain system. Routing is done
pus networks of the ARPA Internet plus state net-
by tables on gateways, which are manually updated
works and a new supercomputer backbone network.
once a month. Modifications have been made to
NSFnet can be viewed as an expansion of the ARPA
sendmail to support routing by domains.
Internet: The ARPANET alone may gain about 15
JUNET has 46 participating organizations and
new PSNs. BITNET is considering migrating from
160 hosts. There are now several links to Europe,
RSCS to TCP/IP, and CSNET is investigating a new
North America, Australia, and Korea. A gateway to
network, Cypress, based on TCP/IP 9,600-bps leased
Asianet (BITNET in Japan) is planned, and there
lines. Both of these may be used as part of NSFnet.
is a gateway to CSNET. The contact address is
The TCP/IP protocols are also being implemented
JUNET under the Cray CTSS operating system.
c/o Jun Murai In addition to services such as already exist in
Computer Center CSNET and the ARPA Internet, it is hoped that
Tokyo Institute of Technology NSFnet will have a sophisticated nameservice ex-
Ookayama Meguroku tending not only to hosts, but also to users and re-
Tokyo 152 Japan sources. A program might be given a networkwide
location-independent name, such as Useful-Program,
which would be mapped by the nameservice into
METANETWORKS
uprog@somedomain. Users would not have to be con-
CSNET cerned with the location of the program.
CSNET is a metanetwork, but has already been de- Long-haul links are expected to typically be Tl
scribed in the section on research networks. microwave links. Reliability will be high due to the
use of TCP/IP. Naming, addressing, and routing will
be as in the ARPA Internet. Administrators of
NSFnet
CSNET, BITNET, and UUCP have recently agreed to
The history of NSFnet properly starts with
adopt the ARPA Internet domain name system
ARPANET and continues with CSNET. In 1984, the
(though conformance is voluntary per host in the
National Science Foundation (NSF) established the
latter two networks). This decision was motivated by
Office of Advanced Scientific Computing (OASC).
purely practical considerations, but will facilitate
That office started one program to develop super-
the implementation of NSFnet.
computer centers, and another to develop network
The network will reach academic campuses and
access to them. The planned network was called
research organizations throughout the United States
NSFnet [42]. This network will eventually provide
and already has 65 initial participant organizations.
more services than just supercomputer access, and
For futher information, contact
serve an extensive user community. Agreement was
reached between NSF and DARPA in October 1985 Program Director for Networking
to provide access to NSFnet for ARPANET users and Office of Advanced Scientific Computing
vice versa. National Science Foundation
The objective of NSFnet is to provide the general Washington, DC 20050

October 1986 Volume 29 Number 10 Communications of the ACM 963

Articles

National Research Internet stages, though there are some existing UUCP links
The National Research Internet (NRI) is being between the various national networks in the Pacific
planned by a network working group of several region, including East Asia, Australia, and the
White House committees. The relevant committees United States.
are collectively known as the Federal Coordinating
Committee on Science, Engineering, and Technology AUSEAnet
(FCCSET), and their purpose is to coordinate activi- AUSEAnet is a metanetwork for a joint microelec-
ties in the various federal agencies. One of them is tronics (VLSI) project among ASEAN (Association of
the FCCSET Committee on Very High Performance South East Asian) countries and Australia. The
Computing (VHPC), which is aimed at the super- ASEAN countries include Thailand, Indonesia, Ma-
computer activities. laysia, Singapore, Brunei, and the Philippines. The
NRI will interconnect the networks owned by the project started in July 1986 and is expected to be
various federal agencies in order to promote re- operational by November 1986. The goal is to permit
source sharing and collaboration among researchers. electronic submission of VLSI designs to the fabrica-
This differs from NSFnet’s purpose in that it at- tion plant in Australia and to exchange information
tempts to satisfy most if not all needs for scientific about microelectronics techniques. Funds are pro-
networking, whereas NSFnet is more directed to- vided by the Australian government and will be
ward providing access to supercomputer centers. augmented by the other participant countries.
NSFnet will be one of the networks connected by AUSEAnet will use UUCP and ACSNET over the
NRI. The NRI project is administered by the international X.25 networks. Outside of Australia,
FCCSET Network Working Group, and the network it will use mostly UUCP. Most of the links will be
will be funded by various federal agencies. Account- 1,200 bps. Reliability should be high.
ing methods are still under study. Country domain and node addressing modes will
The TCP/IP protocol suite will be used for NRI. be decided in August 1986. Indonesia will act as
The main focus of development will be on gateway ASEAN regional center of AUSEAnet and will con-
standards in order to facilitate interconnection of the nect to a designated node in Australia through an
networks. The links used will presumably be fast, international packet-switching line. Each participat-
and the TCP/IP protocols will promote reliability. ing ASEAN country will have its own network and
Access restrictions among the various agency net- an international gateway that will poll the Indone-
works may become an issue, and prior arrangement sian hub machine. The national center in Indonesia
with the administration of a network may be neces- will be at NETLAB, the Network Laboratory that is
sary before a given resource on one network can be part of the Inter-University Center for Computer
used by a user of a different network. Science at the University of Indonesia in Jakarta.
Four institutional organizations were connected to
RARE NETLAB as of July 1986. At least nine ASEAN insti-
The EEC’s ESPRIT project is sponsoring plans for a tutional nodes will participate in AUSEAnet. The liai-
metanetwork called RARE (Reseaux Associees pour son is Joseph F. P. Luhukay, who can be reached at
la Recherche Europeenne). RARE is an attempt to
unify and standardize the European national net- Computer Science Center
works. Funding may come from both the EEC and University of Indonesia
from the governments of participating countries. The P.O. Box 3442
IS0 protocols will be used, though ESPRIT also spon- Jakarta 10002
sors projects li.ke ROSE that may convert existing Indonesia
networks from other protocols gradually. Compo-
nents of some international networks, such as EUnet BULLETIN BOARDS AND NETWORKS
or EARN, may be funded and incorporated. Some Bulletin-board systems have become ubiquitous, and
constituents of this metanetwork already exist in na- many people today think of bulletin-board systems
tional or regional networks like DFN (West Ger- when they hear the word “network.” The two are in
many) or NORDUNET (Scandinavia). NORDUNET fact quite distinct. A typical bulletin-board system is
uses not only IS0 protocols, but also some Coloured a personal computer supporting a database of mes-
Book ones. sages. Users can read or submit messages by dialing
up one or more modems over the public switched
Pacnet telephone network. The telephone network is the
Pacnet is a logical grouping of Pacific hosts and orga- only network involved, since all the messages are on
nizations [g]. It is currently in the early planning one machine.

964 Communications of the ACM October 2986 Volume 29 Number 111

 Articles

Bulletin-board systems are computer conferencing data networks-that are used to reach them become
systems. Unlike mail systems, where communication saturated, and the mainframes themselves become
is one-to-one or one-to-many, communication on loaded.
computer conferencing systems is many-to-many PDNs are real networks, or at least communica-
[32]. People who use personal-computer bulletin- tion subnets, as they are implemented up to the net-
board systems have a tendency to call anything work layer and enable users to reach computers.
larger a network. An example of this is the WELL There are many PDNs; every European country, for
(Whole Earth ‘Lectronic Link) [6], which is a VAX- instance, has a national government-run PDN. The
11/750 with two Fujitsu Eagle disk drives running government agency irtvolved is usually the same one
UNIX (4.2BSD) in the office of the Whole Earth Re- that runs the paper post and telephone services. The
view in Sausalito, California. Despite the difference French acronym PTT (Poste, Tklbphone, et T%-
between the WELL’s 800 megabytes of disk space graphe) is the standard abbreviation for this kind of
and the typical home computer’s few megabytes, the agency.
WELL is still a bulletin-board system. People dial it
up and use it like they would CompuServe or THE Conferencing Networks
SOURCE. The WELL can do more than the average A basic technical difference that distinguishes con-
BBS system, however. There is a local magazine that ferencing systems from mail systems is that confer-
is published on-line, as well as parts of other maga- encing systems need multicast or broadcast transport
zines and books [7]. In addition, the WELL is on mechanisms to support them, whereas mail systems
LJSENET and UUCP and serves as one of the few can get along with unicast mechanisms. This and
public access entry points to those networks. While other technical issues related to the extent and
most USENET users are employees of organizations method of distribution of the message database are
that have machines connected to USENET, anyone not highly visible to the users of any of the BBS or
who can dial up and submit a valid credit-card num- news systems.
ber can (in principle) be a WELL user. Access is not Local bulletih boards have existed for many years
unlimited, however, since the administrators of the on TOPS-20 (and other) machines on the ARPA In-
WELL are aware of the nonprofit nature of USENET. ternet, with one message database per system. Those
There are several other public access UNIX sys- systems pass the messages in the databases around
tems that provide USENET access, such as the Soup to each other so that the database on each system
Kitchen in New Jersey. Few of them support pub- approximates those on the others. The transfer
lishing ventures, however. methods are an ad hoc combination of file transfer of
the message databases and mail of individual mes-
Commercial Networks sages. There are many established ARPANET mail-
Commercial networks sell services to outside users ing lists, many of which are stored on arrival at
for profit. Many are in effect common carriers like TOPS-20 systems as BBOARDS, but these are still
the telephone system. Administration is always cen- not true distributed systems.
tralized, though execution may be delegated. Fees USENET is similar in concept to the TOPS-20
are usually charged to individual persons or organi- BBOARD systems. Each USENET host machine con-
zations on the basis of connect or CPU time used. tains one copy of the news article database. The key
CompuServe, THE SOURCE, and other such difference is that, when a USENET article is posted
services are not really networks. They consist of a locally, it is sent to every neighboring machine,
few large computers closely coupled into a large dis- which in turn sends it to all of its neighbors, and so
tributed system and are accessed just like home forth. To keep this flooding algorithm from produc-
personal-computer bulletin-board systems, except ing loops, each article contains a record of the path
that users get bills. it has traversed, which is updated at each machine
On a commercial service, more traffic is good, be- the article passes through; no machine will send an
cause it brings in more revenue, but, on an anarchic article to a machine that is already in the path. The
network like USENET, more traffic just means more article also contains a unique ID that the hosts use to
expense. Users of either kind of system have to deal recognize and discard duplicates. Many people have
wiih information overload, though the availability taken to having their personal computers dial up
of money and administration in the centralized ser- CompuServe in the middle of the night (when tele-
vices allows the development of sophisticated filter- phone rates are low) and download many articles for
ing mechanisms to limit that problem. But central- later perusal. This is a step in the USENET direction.
ization produces its own problems: During popular Most computer conferencing systems organize
hours, most of the dial-up ports and PDNs-public messages into conferences according to subject mat-

October 1986 Volume 29 Number 10 Communications of the ACM 965

Articles

ter. Many have each conference (SIG, newsgroup, knowledgeable people, and so to keep elementary
etc.) overseen by a person who may be called an questions out of UNIX-WIZARDS. This new news-
editor, a moderator, or a Sysop (the term used on group is popular, at least. There is also a moderated
most commercial and personal systems). This person newsgroup, comp.unix. This has little traffic, appar-
filters out duplicate submissions, and may in some ently because people do not want to have to justify
cases reject ob.jectional submissions or remove them the value of their submissions. UNIX-WIZARDS still
after they are posted. Reasons for rejection vary has a recognizable group of technical contributors
widely according to the network, conference, and and readers who use it in their work. Many of them
people involveId. Sometimes actual editing is done. can also be found attending USENIX conferences for
The role of the moderator-as perceived by the the same reasons. Many of those who no longer fol-
tnoderators themselves, as well as network adminis- low UNIX-WIZARDS use other newsgroups or mail-
trators, submittors, and readers-can vary widely. ing lists or private mail for the same purpose.
For example, accusations of censorship are unfortu-
AILIST@SRI-ALARPA or comp.ai: This is a general
nately common, though few moderators believe they
technical discussion list or newsgroup for artificial
are justified.
intelligence (AI) researchers. It is moderated and di-
SOCIAL AND LEGAL ISSUES gestified. The volume is high, and topics range from
Networks have effects on their users that go beyond press treatment of AI to esoteric points of logic to
the issue of immediate practical utility [32]. The pri- implementation details. Submittors range from the
mary effect is increased human interaction. If infor- most eminent practitioners to novices, with the
mation overload can be avoided, increased interac- moderator selecting more for the former. It is not
tion can lead to better technical productivity clear that this list accurately reflects the working AI
through the exchange of ideas and references. It can community, but it certainly has its own following.
also lead to unanticipated social interaction among [email protected]: This is an ARPA Internet
specialized or diverse groups of people. In addition mailing list that deals with the TCP/IP protocol
to social effects, network interaction has potential suite. It is used both for dissemination of information
legal implications for users and administrators. to people not familiar with the protocols, and for
working technical discussions among their imple-
Networked Communities
mentors, most of whom appear to follow the list.
Certain newsgroups, mailing lists, bulletin boards,
There are other similar lists on more specific net-
and SIGs have reliable followings that form social
working topics.
groups. These range from groups interacting strictly
in pursuit of technical goals to others interacting for news.group: This is a USENET newsgroup for dis-
the sake of interaction, to still others for whom the cussing the creation and deletion of newsgroups. It
networked interaction is an aspect of or leads to has on occasion been one of the highest volume
outside interaction. newsgroups on the network. There are other news-
groups that are also about USENET itself.
Technical Groups.
[email protected]: This ARPA Internet INFO-NETS%[email protected]: This is
mailing list’ dates back to around 1977 on the a mailing list about networks. Typical postings might
ARPANET and i.s currently gatewayed bidirection- deal with requests for paths to specific hosts on cer-
ally and automatically with the USENET newsgroup tain networks or requests for position statements by
comp.unix.wizards. It is possible that most working people involved with NSFnet. Some information in
UNIX software developers and system administra- this article was obtained from responses to requests
tors read this list up to a few years ago, but many on this list.
have since canceled their subscriptions because of [email protected]: HUMAN-
how long it takes to sort through the much larger NETS is perhaps the prototypical technical list about
volume of submissions. There have been several social issues. It is a forum for discussions of the so-
attempts to reduce the traffic and to keep it more cial effects of computers and specifically of com-
technical. The comp.unix.help newsgroup, which is puter networks. A discussion in this list led to the
gatewayed with the [email protected] mailing writing of an earlier version of this article.
list, was created to provide novices with access to
There are technical mailing lists for such things as
>The convention for subscribing to ARPA Internet mailing lists is to send mail
to list-REQUEST@domain. not list@domain. For example: If you want to get workstations, local-area networks, and many differ-
on the HUMAN-NETS list. mail a request to HUMAN-NETS-REQLJEST@- ent lists for many different manufacturers’ com-
RED.RUTGERS.EDU. Only actual submissions should go to HUMAN-
[email protected]:DU. puters. Not all technical lists or newsgroups are

966 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

computer related. There are newsgroups on astron- Legal Issues

omy and biology, for instance. However, researchers The specific liabilities that arise when computers
in other fields do not use newsgroups in their fields communicate with other computers over public air-
for actual work as much as researchers in computer ways or through the telephone system can be diffi-
fields do, probably because researchers in other cult to recognize. The extent of a system’s liability is
fields are less familiar with unusual uses of com- a function of its specific features and of the extent to
puters. which the content of its messages is controlled.
There are legal precedents covering the liabilities of
Social Groups. more traditional communication media such as
CompuServe SF SIG: CompuServe has a very popu- newspapers, radio and television broadcasting, and
lar though not very old special interest group (SIG) cable television. Though computer networks do not
on science fiction, moderated in part by Diane fit neatly into either the broadcaster or common-
Duanes, a popular science-fiction and fantasy writer. carrier category, these are the two classifications
The instigators had great difficulty convincing net- that seem likely to provide legal precedents that will
work management that a conference on this topic apply to computer communications [78]-the alter-
would be viable, but it has turned out to be one of native is to define a special classification for com-
the fastest growing SIGs. Similar groups exist on puter communications. The classification is impor-
some other systems. tant because common carriers are not held to as
strict a standard as broadcasters. Some of the liabili-
There are many newsgroups or mailing lists that ties faced by network administrators have to do with
exist only for social purposes. A famous example defamatory material, obscenity, content of transmis-
among aficionados was a mailing list started by a sion, and faulty transmission. Individual users might
student who had lost his girlfriend and wanted to also be liable for defamatory material and obscenity,
commiserate with all his friends, most of whom he as well as for copyright infringement and invasion of
knew through the various networks. This list used privacy.
considerable portions of the bandwidth of several To control what is posted on a network, it is nec-
networks over many months and led to a number of essary to control access to that network. Most exist-
parties in several parts of North America where the ing networks are not strong on security. The safest
participants met each other directly. This list was policy in using networks is to assume that any net-
never sanctioned by the administrators of any net- work can be broken, that any transmission can be
work. recorded, and that most can be forged. (There was a
famous hoax on April Fools’ Day in 1984 when
Social Effects kremvax!kgbvax!chernenko joined USENET.) En-
One of the most obvious effects of networks is their cryption techniques for providing a rather high de-
tendency to induce users to “flame,” that is, to pro- gree of security exist, but few people are willing to
duce many words on an uninteresting topic, or in an pay the price in CPU time, and so few networks
abusive or ridiculous manner; “raving” is almost a make use of these techniques.
synonym for flaming. The usual explanation for why
computer networks tend to aggravate flaming is that HISTORY
the flamer is isolated from the readers and has no The first packet-switching network was imple-
negative feedback to inhibit such behavior. mented at the National Physical Laboratories in
There are typographic conventions that have de- the United Kingdom. It was quickly followed by the
veloped on the various networks to get around the ARPANET in 1969 [15]. There were soon related
difficulties of expressing nuances in ASCII charac- projects such as CYCLADES in France [72], EIN
ters. One of the more universal is that UPPERCASE (European Informatics Network) in Europe [2, 211,
means shouting (much to the chagrin of those with and the Coloured Book efforts in the United Kingdom.
micros that only have uppercase). Some *surround Of these, the networks based on the ARPANET and
phrases with asterisks* to indicate emphasis, while Coloured Book protocols have survived, by mutation
others s p a c e the characters out. People will mark and evolution of their protocols. These original ef-
(sarcasm) or (irony). Facial expressions often get forts were all undertaken as research and were
spelled out (*grin*). There are many ways to indi- largely government supported.
cate the start of a flame, such as *FLAME ON!*. On As the idea of networking caught on, companies
USENET there are shorter ways to indicate lack of such as Xerox, DEC, and IBM started to develop
serious intent: for instance, :-) (the image of their own networking technologies, usually starting
a smiling face). with local-area networks. Long-haul networks came

October 1986 Volume 29 Number 10 Communications of the ACM 967

 Articles

to be used not only for communication among di- have many workstations on their LAN components,
rectly connected hosts, but also to tie LANs into so the average size of their hosts is smaller. Personal
internets. computers are sometimes connected to internets like
ARPANET technology had been used by BBN to the ARPA Internet, and some are part of the dial-up
build the commercial network TELENET (later sold networks. Users of IBM PCs have found a network of
to GTE) by 1976, and commercial X.25-based net- their own in FidoNet. At the other end of the spec-
works followed. In Europe the PTTs controlled (and trum, at least one network, MFENET, was developed
still control) the PDNs in each country (one per primarily for access to supercomputers.
country) and hlave universally settled on X.25 as Computer networks have spread to larger and
their network layer protocol. The PTTs favor circuit smaller machines, different lower layer technologies,
switching rather than packet switching, so most of different protocols, and many nations. Though their
the CCITT protocols such as X.25 and X.400 are ori- diversity continues to increase, most noncommercial
ented toward virtual circuits. networks are connected with each other at least for
Computer conferencing systems started in 1976 the purposes of mail exchange and thus already con-
and later found commercial viability in centralized stitute a worldwide metanetwork, first predicted
services such as EIES and Delphi as well as some- years ago and called Worldnet (see Figure 2).
what more distributed systems like CompuServe.
Personal computers are often used as free bulletin BIBLIOGRAPHIC NOTES
boards. A good introduction to actual functioning networks
Meanwhile, another networking technology, based is [50]. The standard tutorial introduction to the the-
on dial-up telephone links instead of dedicated con- ory of computer networks is still [85]. The reader
nections, was being developed. Two of the earliest can gain a useful amount of context by reading just
products of thi,s technology were ACSNET and its first chapter, and there is an overview paper by
UUCP, both of which survive in modified forms. The the same author [86]. Either of these should be sup-
dial-up networks produced the most distributed of plemented by recent publications on the various
the conferencing systems: USENET. protocols and protocol suites, for which see the
CSNET started as an attempt to bring the collabo- specific references. A perspective on historical and
rative advantages of the ARPANET to researchers recent developments in protocol design and imple-
beyond the ARPANET community by using dial-up mentation can be found in [66]. Good discussions
mechanisms similar to those of UUCP. MAILNET concerning networks and protocols can be found
grew out of a similar effort. BITNET made IBM’s in [38], and there are some useful comparative net-
internal mainframe networking technology available work papers in [13]. Much of the information in this
to the academic community and even spread to article was obtained over the networks it describes,
some non-IBM hosts. It has also spread outside the either by mail, file transfer, or news. Some of the
United States as NETNORTH and EARN. references are to articles in well-known digests
Internets required new protocol suites, such as or newsgroups.
Xerox NS, the U.S. DOD’S TCP/IP, and the IS0 pro-
tocols. The spread of NS has, some say, been stifled Acknowledgments. The number of people who
by the secrecy of its originating company. The have supplied information for or otherwise helped
TCP/IP protocols are by far the most widely imple- with this article is much too large for their names to
mented of these three due to the accessibility of be listed here; the authors thank all of them. How-
their specifications, their long history of practical ever, we would like to especially acknowledge a few
use, and the backing of the U.S. government. Some whose aid has extended far beyond their own net-
of the IS0 protocols found implementation in 1983 works and the call of duty, including Rick Adams,
on CDNnet, the first EAN network, and spread rap- Janet Asteroff, Piet Beertema, Kilnam Chon, Pete
idly in Europe the following year. Other implemen- Collinson, John R. Covert, Franklin Davis, Peter J.
tations, particularly of X.400 (actually a CCITT pro- Denning, Robert Elz, Erik Fair, Michel Gien, Mark
tocol), have followed, especially in Europe. Most of Horton, Jin H. Hur, Christian Huitema, Daniel
the IS0 protocols are either adoptions of CCITT Karrenberg, Steve Kille, Joshua Knight, John Larson,
protocols or are, like them, oriented toward virtual Barry Leiner, Jun Murai, Henry Nussbacher, Craig
circuits. Partridge, Hans Strack-Zimmermann, Gligor
Hosts on early networks were usually either main- Tashkovich, Valerie Thomas, Hide Tokuda, and
frames or minicomputers. A few networks, such as doubtless others whom we have inadvertently ne-
BITNET, continue this tradition. Internets usually glected to mention. Many network administrators

966 Communications of the ACM October 1986 Volume 29 Number 10

1969 1970 1972 1974 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

CYCLADES

MILNET
*
ARPANET ARPA Irlferwf DDN
-\ *
‘\ \ NSFnet
\ ;-
\
\ CSNET /’
\ \ *
\ \ \ Cypress
\ \
\ *
\ ‘, X25NET
\ *
\
\ Phonenet
*

SERCnet JANET
*

CDNnet
b
European EAN networks

---_--_-- COSAC
*
SMARTIX

Xerox RIN Xerox Internet

*
Xerox CIN
*

DECNET E-net Easynet

‘\\ SPAN
.

VNET
\ *
\ NETNORTH
\ *
\ BITNET
*
EARN
*

FidoNet

ACSNET
\ *
\
\
EUnet \
>
USENET
*
UUCP
*
SDN A
JUNET
*

1969 1970 1972 1974 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

Solid lines indicate a continuing operational network, and dashed lines indicate a transfer
of technology to a newer network. Italics indicate internets.

FIGURE2. Time Lines for the Development of Notable Computer Networks

October 7986 Volume 29 Number 10 Communications of the ACM 969

 Articles

have supplied information, but information included 25. Gilmore, P., and Neufeld, G. A strategy for a national electronic
messaging system for research in Canada. In Proceedings of the Cana-
about any network may not necessarily reflect the dian Information Processing Society National Meeting (Montreal, Can-
views of its administrators, users, or others associ- ada, June). 1985.
26. Ginsberg, K. Getting from here to there. In Unix Review, vol. 4, no. I,
ated with it. Inclusion or omission of a section on M. Compton. Ed. Freeman, San Francisco, Calif., Jan. 1986, p. 45.
any network d.oes not imply anything about the im- (“How UNIX communications facilities compare with the routing
mechanisms of TYMNET, DNA, SNA, and ARPANET-and why
portance, quality, size, or an.y other property of that UNIX users should care.“)
network: The networks included are simply those 27. Green, J.L., and King, J.H. Behind the scenes during a comet encoun-
ter. EOS 67, 105 (Mar. 1986).
that the authors noticed. Any intentional inaccura- 28. Green, J.L., and Peters, D., Eds. Introduction to the space physics
cies or other falults in this article are the sole respon- analysis network (SPAN). TM-86499, NASA Ames Research Center,
Moffett Field, Calif.. Apr. 1985.
sibility of the aluthors. 29. Green, J.L., and Zwickl, R.D. Data system users working group meet-
ing report. EOS (meeting report) 67, 8 (1986).
30. Green, J.L., Peters, D.J., Heijden, N., and Lopez-Swafford, B. Manage-
ment of the space physics analysis network. NSSDC Publication,
REFERENCES July 1986.
1. Allman, E. Interview with Peter Honeyman. In UNIX Review, vol. 4, 31. Hendricks, E.C., and Hartman, T.C. Evolution of a virtual machine
no. 1, M. Compton. Ed. Freeman, San Francisco, Calif., Jan. 1986, subsystem. IBM Syst. 1. 18, 1 (1979), 111.
p. 64. 32. Hiltz, S.R., and Turoff, M. Structuring computer-mediated commu-
2. Barber, D.L.A. A European informatics network: Achievement and nication systems to avoid information overload. Commun. ACM 28, 7
prospects. In IEEE ‘76 (Toronto. Canada, Aug.). IEEE, New York, (July 1985), 680-689.
1976, pp. 44-50, 33. Hinden, R., and Sheltzer, A. The DARPA internet gateway; RFC823.
3. BBN Laboratories. 4th anniversary issue. CSNET News 8 [Summer In ARPANET Working Group Requests for Comments, no. 823. SRI
1985). CSNET CIC, BBN Laboratories, Cambridge, Mass. International, Menlo Park, Calif., Sept. 1982.
4. Blumann, W., Cadwallader, R.. Diediw, A., Lovelock, J., Power. D., 34. Horton, M.R. Standard for interchange of USENET messages;
Pozzana, S., and Saury, C. Implementation of OS1 protocols in the RFC850. In ARPANET Working Group Requests for Comments, no. 850.
ESPRIT information exchange system. In ESPRIT ‘85: Status Report of SRI International, Menlo Park, Calif., June 1983.
Confinuing Work. Elsevier North-Holland, New York, 1986, pp. 1387- 35. Horton, M.R. What is a domain? In USENIX Association Conference
1396. Proceedings (Salt Lake City, Utah, June 12-15). USENIX Association,
5. Boggs, D.R., et al. Pup: An internetwork architecture. IEEE Trans. El Cerrito, Calif.. 1984, pp. 368-372.
Commun. COM-28.4 (1980), 612-624. 36. Horton, M.R. UUCP mail interchange format standard: RFC976. In
6. 13rand, S. Gate Five Road. Whole Earth Rev. 47 (July 1985), 103-104. ARPANET Working Group Requests for Comments, no. 976. SRI Inter-
7. Brand, S. Gate Five Road. Whole Earth Rev. 48 (Fall 1985), 139. national, Menlo Park, Calif., Feb. 1986.
8. Cerf, V.G., and Cain, E. The DOD internet architecture model. 37. Horton, M.R.. Summers-Horton, K., and Kercheval, B. Proposal for a
Comput. Networks 7 (1983). 307-318. UUCP/Usenet registry host. In USENIX Association Conference Pro-
9. Chon, K. National and regional computer networks for academic ceedings (Salt Lake City, Utah, June 12-15). USENlX Association,
and research communities in the Pacific region. In Proceedings of El Cerrito, Calif., 1984, p. 373.
PCCS (Seoul. Korea. Oct.). 1985, pp. 560-566. 38. IEEE. Proceedings of the 25th IEEE Computer Society International Con-
10. Chon. K., et al. System development network. In Proceedings of ference (COMPCONJ. Society Press, Los Angeles, Calif., Sept. 1982.
TENCON (Singapore, Apr.). 1984, pp. 133-135. 39. International Organization for Standardization. IS0 open systems
11. Chon. K., et al. SDN: A computer network for Korean research interconnection-Basic reference model. ISO/TC 97/SC, vol. 16, no.
community. In Proceedings ofPCCS (Seoul, Korea. Oct.). 1985, 719, International Organization for Standardization. Aug. 1981. Or-
pp. 567-570. dering information: American National Standards Institute, 1430
12. Comer, D. The computer science research network CSNET: A his- Broadway, New York, NY 10018. Phone: l-212-642-4900; elsewhere
tory and status report. Commun. ACM 26, 10 (Oct. 1983), 747-753. there are equivalent national ordering offices.
13. Compton, M.. Ed. Unix Review, vol. 4. no. 1. Freeman, San Francisco, 40. Ishida, H. Current status of the N-l inter-university network with
Calif., Jan. 1986. access to supercomputers in Japan. In Proceedings of PCCS (Seoul,
14. Cracker, D.H., Viltal, J.J., Pogran, K.T., and Henderson, D.A., Jr. Korea, Oct.). 1985.
Standard for the format of ARPA network text messages; RFC733. In 41. Jacobsen, O., and Postel, J. Protocol document order information;
ARPANET Workin‘g Group Requests for Comments. SRI International, RFC980. In ARPANET Working Group Requests for Comments, no. 980.
Menlo Park, Calif.. Nov. 21.1977. SRI International, Menlo Park, Calif., Mar. 1986.
15. DARPA. A History of the ARPANET: The First Decade. Bolt, Beranek 42. Jennings, D.M., Landweber. L.H., Fuchs, I.H., Farber, D.J., and
and Newman, Cambridge, Mass., Apr. 1983. (Defense Tech. Info. Ctr. Adrion. W.R. Computer networking for scientists. Science 231 (Feb.
AD Al 15440.) 28,1986). 943-950.
16. Davis, F. “Worldnet” and X.400 in Switzerland. IAM, Universitaet 43. Kawaguchi, K., Sate, K., Sample, R., Demco, J,, and Hilpert, B. Inter-
Bern, Bern. Switzerland. Apr. 1986. i’[email protected]~ connecting two X.400 message systems. In Proceedings $ the 2nd
17. Dennett, S., Feinler, E.J., and Perillo, F. ARPANET Znformation Bro- International Symposium on Computer Message Systems (Sept.). IFIP,
chure. DDN Network Information Center, SRI International, Room Washington, D.C., 1985, pp. 15-26. (Also in Computer Message Sys-
EJ291, 333 Ravenswood Ave., Menlo Park, CA 94025, Dec. 1985. tems-85, pp. 17-28.)
18. Denning, P.J. Internal structure of networks. Am. Sri. (Mar.-Apr. 44. Kiessig. R. UUCP Network Directory, vol. 2, no. 2. Spring 1986.
1985). P.O. Box 50174, Palo Alto, CA 94303.
19. Denning, P.J. Supernetworks. Am. Sci. (May-June 1985), 225-227. 45. Kille, SE. Mapping between X.400 and RFC822; RFC987. In
20. Denning. P.J., Hear”, A.C., and Kern, C.W. History and overview of ARPANET Working Group Requests for Comments, no. 987. SRI Inter-
CSNET. In Proceedings of the ACM SIGCOMM Symposium (Mar.). national, Menlo Park, Calif., June 1986.
ACM, New York, 1983, pp. 138-145. 46. Kluger. L., and Shoch, J. Names, addresses, and routes. In Unix
21. Deparis, M., Duenki, A., Gien. M.. Louis, J., LeMoli, G., and Weak- Review, vol. 4, no. 1, M. Compton, Ed. Freeman, San Francisco,
ing, K. The implementation of an end-to-end protocol by EIN Calif., Jan. 1986, p. 30.
centres: A survey and comparison. In Proceedings of IEEE ‘76 (To- 47. Kummerfeld, R.J. ACSnet Current Stafus and Future Developmenf. Ko-
ronto, Canada, Aug.). IEEE, New York, 1976. rea, 1985.
22. Dick-Lauder, P., Kummerfeld, R.J., and Elz, R. ACSNET-The Aus- 48. Kummerfeld, R.J.. and Dick-Lauder, P.R. The Sydney Unix network.
tralian alternative to UUCP. In USENIX Association Conference Pro- Aust. Comput. J. 13, 2 (May 1981), 52-57.
ceedings (Salt Lake City, Utah, June 12-15). USENIX Association, El 49. Kummerfeld, R.J., and Dick-Lauder, P.R. Domain addressing in SUN
Cerrito, Calif.. 1984. pp. 11-17. III. In Proceedings of EUUG (Paris). 1985.
23. Emerson, S.L. LJSENET: A bulletin board for UNIX users. Byte (Oct. 50. Landweber, L.H., Jennings, D.M., and Fuchs, I. Research computer
19831, 219. networks and their interconnection. 1EEE Commun. Msg. 24, 6 (June
24. Fuchs, I.H. BITNET: Because it’s time. Perspect. in Comput. 3, 1 (Mar. 1986), 5-17.
1983). 51. Leiner, B.M., Cole, R.. Postel, J.. and Mills, D. The DARPA internet

970 Communications of the ACM October 1986 Volume 29 Number 10

 Articles

protocol suite. IEEE Commun. Mug. (Mar. 1985). [Also in 1985 DDN 76. Seamonson, L.J., and Rosen, E.C. “STUB” Exterior gateway protocol;
Protocol Handbook, vol. 2.) RFC888. In ARPANET Working Group Requests for Comments, no. 888.
52. McQuillan, J.M., and Walden, D.C. The ARPA network design deci- SRI International, Menlo Park, Calif., Jan. 1984.
sions. Comput. Networks 1 (1977), 243-269. 77. Shoch, J.F. Internetwork naming, addressing, and routing. In Pro-
53. MIL-STD. Military Standards for DOD Internet Protocols. Naval Publi- ceedings of the 17th IEEE Computer Society Infernational Conference
cations and Forms Center, Code 3015, 5801 Tabor Ave., Philadel- (COMPCOM (Sept.). IEEE, New York, 1978, pp. 430-437.
phia, PA 19120. The ARPA Internet protocols are defined by the set 78. Shulman, C.H. Legal research on USENET liability issues. ;login: The
of military standards IP (MIL-STD-1777), TCP (MIL-STD-1778). FTP USENIX Assoc. News/. 9, 6 (Dec. 1984), 11-17. (USENIX Association,
(MIL-STD-1780). SMTP (MIL-STD-1781), and TELNET (MIL-STD- El Cerrito, Calif.)
1782). See also ARPANET Working Group Requests for Comments and 79. Simonsen, K.J. Re: Re: Network differences. net.unix, no.
1985 DDN Protocol Handbook. ([email protected]), USENET, Dec. 18,1985.
54. Mills, D.L. Exterior gateway protocol formal specification; RFC904. 80. SPAG. Guide to the Use of Standards. 1985.
In ARPANET Working Group Requests for Comments, no. 904. SRI 81. Spratt, E.B. Networking developments in the U.K. academic commu-
International, Menlo Park, Calif., Apr. 1984. nity. In International Conference on Information Network and Data
55. Mockapetris, P. Domain names-Concepts and facilities; RFC882. In Communication (INDC-86), IFIP TC.6 (Ronnedy, Sweden, May). IFIP,
ARPANET Working Group Requests for Comments, no. 882. SRI Inter- Washington, DC., 1986.
national, Menlo Park, Calif., Nov. 1983. 82. SRI International. 1985 DDN Protocol Handbook. DDN Network Infor-
56. Mockapetris, P. Domain names-Implementation and specification: mation Center, SRI International, Room EJ291, 333 Ravenswood
RFC883. In ARPANET Working Group Requests for Comments, no. 883. Ave., Menlo Park, CA 94025, 1985. Most relevant papers from both
SRI International, Menlo Park, Calif., Nov. 1983. ARPANET Working Group Requests for Comments and Military Stand-
57. Mockapetris, P. The domain name system. In Proceedings of the IFJP ards for DOD Internet Protocols are collected here.
6.5 Working Conference on Computer Message Services (Nottingham, 83. SRI International. ARPANET Working Group Requests for Comments.
England, May). IFIP. Washington, DC., 1984. (Also as ISI/RS-84-133, DDN Network Information Center, SRI International, Room EJ291,
June 1984.) 333 Ravenswood Ave., Menlo Park, CA 94025. This series of techni-
58. Mockapetris, P. Domain system changes and observations: RFC973. cal notes includes the specifications for the ARPA Internet protocols
In ARPANET Working Group Requests for Comments, no. 973. SRI IP (RFC791), ICMP (RFC792], TCP (RFC793), UDP (RFC768), FTP
International, Menlo Park, Calif., Jan. 1986. (RFC959). SMTP (RFC821), and TELNET (RFC854), plus related
59. Mockapetris, P., Pastel. J.. and Kirton, P. Name server design for papers. All the protocols are indexed in Assigned Numbers
distributed systems. In Proceedings of the 7th International Conference (RFC960) and Official ARPA-Internet Protocols (RFC961).
on Computer Communication (Sydney, Australia, Oct.). 1984. (Also as 84. Summers-Horton, K., and Horton, M. Status of the USENIX UUCP
ISl/RS-84-132, June 1984.) project. In USENIX Association Conference Proceedings (Dallas, Tex.,
60. Mogul, J., and Pastel, J. Internet standard subnetting procedure: Jan. 23-25). USENIX Association, El Cerrito. Calif.. 1985, p. 183.
RFC950. In ARPANET Working Group Requests for Comments, no. 950. 85. Tanenbaum. AS Computer Networks. Prentice-Hall Software Series,
SRI International, Menlo Park, Calif., Aug. 1985. Prentice-Hall, Englewood Cliffs, N.J., 1981.
61. Murai, J.. and Asami. T. A network for research and development 86. Tanenbaum. A.S. Network protocols. ACM Comput. Sure. 13, 4 (Dec.
communities in Japan-JUNET. In Proceedings of PCCS (Kaist, Korea, 1981). 453-489.
Oct.). 1985, pp. 579-588. 87. Uhlig, R., Ed. Computer Message Systems-85. Elsevier North-
62. Neufeld, G., Demco. J., Hilpert, B., and Sample, R. EAN: An X.400 Holland, New York, 1986.
message system. In Proceedings of the 2nd Infernafional Symposium on 88. VNET-BITNET Gateway General User’s Guide. Available on BITNET
Computer Message Systems (Sept.). IFIP, Washington, D.C.. 1985, via INFO@BITNIC, Mar. 22, 1985.
pp. l-13. (Also in Cornpurer Message Systems-85, pp. 3-15.) 89. Weinstein, L. Project Stargate. In USENlX Association Conference
63. Nowitz, D.A. Uucp implementation description. In UNIX Program- Proceedings (Portland, Oreg., June 11-14). USENIX Association,
mer’s Manual, 7th ed., vol. 2. Holt, Rinehart and Winston, New York, El Cerrito. Calif.. 1985. pp. 79-80.
1983. 90. Weinstein, L. Project Stargate. In EUUG Conference Proceedings
64. Nowitz, D.A., and Lesk, M.E. A dial-up network of UNIX systems. In (Florence, Italy, Apr.). European Unix Systems Users’ Group,
UNlX Programmer’s Manual, 7th ed., vol. 2. Halt, Rinehart and Win- Buntingford, England, 1986.
ston, New York, 1983. 91. Xerox Corporation. The Ethernet, a local area network: Data link
65. Organick, E.I. The Multics System: An Examinarion of Its Structure. layer and physical layer specification. X3T51/80-50. Xerox Corpora-
MIT Press, Cambridge, Mass., 1975. tion, Stamford, Corm., Oct. 1980.
66. Padlipsky, M.A. The Elements of Networking Style. Prentice-Hall, En- 92. Xerox Corporation. An internetwork architecture. XSIS 028112,
glewood Cliffs, N.J., 1985. Xerox Corporation, Stamford, Corm., Dec. 1981.
67. Partridge, C. Mail routing and the domain system; RFC974. In
ARPANET Working Group Requests for Comments, no. 974. SRI Inter-
national, Menlo Park, Calif., Jan. 1986.
68. Partridge, C. Report from the Internet NIC on domains, CSNET-
FORUM Dig. 2, 2 (Feb. 19, 1986). CSNET CIC. BBN Laboratories, CR Categories and Subject Descriptors: C.2.0 [Computer-
Cambridge, Mass. Communication Networks]: General-data communicofions: K.2 [History
69. Partridge, C. Mail routing using domain names: An informal tour. In of Computing]: systems, networks
Proceedings of the 1986 Summer USENIX Conference (Atlanta, Ga., General Terms: Management, Performance, Reliability, Standardiza-
June 9-13). USENIX Association, El Cerrito, Calif., 1986, pp. 366- tion
376. Additional Key Words and Phrases: Addressing, domains, gateways,
70. Postel, J. Domain name system implementation schedule-Revised; networks, protocols, Worldnet
RFCSPI. In ARPANET Working Group Requests for Comments, no. 921.
SRI International, Menlo Park, Calif., Oct. 1984.
71. Pastel. J., and Reynolds, J, Domain requirements; RFC920. In
ARPANET Working Group Requests for Comments, no. 920. SRI Inter-
national, Menlo Park, Calif., Oct. 1984.
72. Pouzin, L., Ed. The CYCLADES computer network-Towards lay- Author’s Present Address: John S. Quarterman. P.O. Box 14621, Austin,
ered network architectures. In Monograph Series of the ICCC, vol. 2. TX 78761; Josiah C. Hoskins, MCC, 9430 Research Blvd.. Kaleido
Elsevier North-Holland, New York, 1982, p. 387. Bldg. 2, Austin, TX 78759-6509.
73. Quarterman, J.S., Silberschatz, A., and Peterson, J.L. 4.2BSD and
4.3BSD as examples of the UNIX system. ACM Comput. Surv. I7,4
(Dec. 1985), 379-418.
74. Ritchie, D.M., and Thompson, K. The UNIX time-sharing system.
Bell Syst. Tech. 1. 57. 6, Part 2 (July-Aug. 1978), 1905-1929. (The Permission to copy without fee all or part of this material is granted
original version (Commun. ACM 17, 7 [July 19741, 365-375.) de- provided that the copies are not made or distributed for direct commer-
scribed Version 6. whereas this one describes Version 7.) cial advantage, the ACM copyright notice and the title of the publication
75. Sanderson, T., Ho, S., Heijden, N., Jabs, E.. and Green, J.L. Near- and its date appear, and notice is given that copying is by permission of
realtime data transmission during the ICE-comet Giacobini-Zinner the Association for Computing Machinery. To copy otherwise, or to
encounter. ESA Bull. 45, 21 (1986). republish, requires a fee and/or specific permission.

October 1986 Volume 29 Number 10 Communications of the ACM 971