HISTORY AND OVERVIEW OF CSNET
Peter J. Denning
C. William Kern
bstract: CSNET is the acronym for the Computer Science Research
Network, a project supported by NSF to provide advanced computer
etwork services to the computer research community. CSNET is a
"logical net" -- a high-level communication environment spanning
everal physical nets, including the ARPANET, Phonenet, and
Telenet. This paper reviews the history, the goals, the organization
he components, and the status of CSNET.
Authors addresses: Peter J. Denning, Computer Sciences Department, Purdue University, West Lafayette, IN 47907;
Anthony Hearn, The Rand Corporation, Santa Monica, CA 90406; C. William Kern, Computer Science Section, Na
ional Science Foundation, Washington, DC 20550. The work reported herein was supported through NSF by awards
to Purdue University, The Rand Corporation, The University of Delaware, and the University of Wisconsin.
This is a preprint of a paper to be presented in a session about the CSNET Project at the ACM SIGCOMM symposi-
um on data communications, March 8-9, 1983.
ACM SIGCOMM, March 1983, 138-145.
CSNET Overview (12/1/82)
SNET stands for the Computer Science Research Network, a project sponsored by NSF to pro-
vide advanced computer network services to the computer research community. CSNET is a "logical
et" -- a high-level communication environment spanning several physical nets, including the
ARPANET, Phonenet, and X.25 public packet-switched networks (e.g., Telenet). This paper review
he history, the goals, the organization, the components, and the status of CSNET as of Fall 1982.
The seeds of CSNET were planted in May 1979. L. H. Landweber arranged a special meeting at
the University of Wisconsin to discuss how computer network services like those of ARPANET could
ecome available to the entire community of computer science researchers. The ARPANET served
only a dozen university sites and DARPA was unable to include the rest of the community.
It was clear to the participants that mail, ﬁle transfer, and remote login services had greatly
enhanced research productivity and had generated a strong community spirit among ARPANET sites. I
as also clear that the ARPANET experiment had produced a split between the "haves" of the
ARPANET and the "have-nots" of the rest of the computer science community. The participants at th
eeting wanted to unify the computer research community and to improve research conditions for all
Each participant at the meeting had previous, favorable experience with computer-based mail ser-
vices for a small research community. One was "THEORYNET", a mailbox facility on a machine at
More detailed information on the background and history of CSNET can be found in the article by Comer . More
detailed information about CSNET’s multinet architecture can be found in the paper by Landweber and Solomon .
CSNET Overview (12/1/82)
he University of Wisconsin accessible via Telenet login to some 200 theoretical computer scientists.
Another was "SAMNET", a mailbox facility on a machine at the University of Toronto accessible to
ome 50 performance analysts. A third was "SYMBOLNET", a network linking four sites comprising
researchers in symbolic computation; the goal was a network based on Telenet with services to mirro
RPANET (but to be independent of ARPANET). Each of these primitive electronic mail services had
signiﬁcantly aided its special community. Each community was anxious to obtain the more advanced
ervices of machine-to-machine mail transfer, ﬁle transfer, and remote login.
Also present at this meeting were observers from NSF and DARPA. The NSF was the sponsor o
HEORYNET and SYMBOLNET and, on the advice of its advisory panel, had been alert for possible
extensions of network services to the rest of the community. DARPA was interested in furthering com
uter science research and believed that network services might be a step of high leverage. These
observers offered advice and encouragement.
The group determined to submit a proposal to the NSF. They envisaged a network available to
all members of the computer science community. The network would have low entry costs and othe
osts proportional to usage. Public packet-switched networks, notably Telenet, would be the underlying
medium. (At the time, a DARPA IMP cost about $90K per year, a price well beyond the means of
ost computer science departments. Moderate speed Telenet connections were then available for entry
fees of about $20K per year.)
In November 1979 the University of Wisconsin submitted a proposal to NSF containing the above
plan on behalf of a consortium of universities (Georgia Tech, Minnesota, New Mexico, Oklahoma, Pur-
ue, UC-Berkeley, Utah, Virginia, Washington, Wisconsin, and Yale). The cost would be $3 million
over a ﬁve-year period. The NSF had the proposal reviewed and returned comments to the proposers i
The reviews revealed a much higher level of skepticism than the proposers had anticipated. The
skepticism had three roots. One was a belief that the CSNET project was proposing to reinvent
CSNET Overview (12/1/82
RPANET technology; this perception was reinforced by the lack of any gateway between the
ARPANET and the proposed CSNET. The second was a belief that insufﬁcient attention had bee
iven to project management and task distribution. The third was a belief that NSF might have to
obtain funds for CSNET by reducing basic research budgets. On the basis of these reviews the NS
ould not fund the project as proposed.
But the NSF’s advisors and some of the reviewers adamantly maintained that CSNET would so
improve research productivity that the potential diversion of funds from basic research would pay off
andsomely in the long run. Accordingly, the NSF offered to fund a thorough study of the CSNET
concept. The purpose of the study was to determine the most cost-effective architecture, to develop
ound management plan, and to assess the extent of community support. With sufﬁciently strong peer
approval, CSNET would be possible.
During the summer of 1980, Landweber convened a CSNET planning committee comprising
nineteen computer scientists, a cross section of leaders of the community who had extensive compute
etwork experience. Two major new factors entered the discussion. One was the existence of MMDF,
software for "multi-channel memo distribution facility," under development at the University of
elaware . MMDF is a UNIX-based mail transport system that sends and receives mail over a
variety of channels including ARPANET and telephone. (The latter is functionally similar to the
uucp" facility of UNIX.) With MMDF, CSNET could bring a large number of sites online in a short
period at low cost.
The second factor was DARPA’s decision to proceed with "internet protocols" (IP) that permit a
host in one net to communicate with a host in a different net. With the internet protocols, CSNET
ould be regarded as a logical organization of users on different nets. DARPA offered to make its new
protocol software (TCP/IP) available to the CSNET project. In return, DARPA expected that the exist
ng ARPANET university community would be a component of CSNET at no cost to DARPA.
nThe planning committee quickly reached a consensus. CSNET would include subnets based o
CSNET Overview (12/1/82
RPANET, X.25 nets, and Phonenet (the MMDF service). The internet protocols would hide these
components from their users. CSNET would develop an interface between ARPANET’s protocol
oftware (IP) and the X.25 public networks (initially Telenet); this would make the standard ARPANET
services available to non-DARPA hosts. CSNET would provide a name server that registers all CSNE
sers and quickly locates the mailbox of any registered user. CSNET would initially receive full sup-
port from NSF, but would become self supporting within ﬁve years via dues and usage fees.
A proposal containing this plan was submitted by Wisconsin on behalf of a consortium of institu-
tions (Wisconsin, Purdue, Utah, Delaware, and The Rand Corporation) in November, 1980. It was
eviewed by late December, 1980 and then submitted to the National Science Board, which approved
the project in January 1981. The Board stipulated that the NSF would provide a full time project
anager for the ﬁrst two years but would withdraw from project management by February, 1983, when
the CSNET organization would be strong enough to take over. Contracts for CSNET were let to
isconsin, Purdue, Delaware, and Rand in late spring, 1981. After 20 months, the seeds of CSNET
had begun to sprout. The real work lay ahead.
The goals of CSNET are summarized in Figure 1. The net is to be open to all computer
researchers throughout the United States (later, the world). It is to be self-supporting. Its users wil
ay ﬁxed annual dues plus usage fees. (The dues for 1983 are shown in the bottom part of the table.)
CSNET will initially comprise three subnets -- ARPANET, Telenet, and Phonenet -- but will be
xpandable to others as they become available, e.g., other X.25 public nets and satellite nets. CSNET
will initially provide the same services as ARPANET -- mail, ﬁle transfer, remote login, and an on-lin
ame server. Later, it will provide additional services such as messages containing voice segments,
software libraries, technical report repositories, and an on-line journal.
-6- )CSNET Overview (12/1/82
COMPUTER SCIENCE RESEARCH NETWORK GOALS
Open to all computer researchers.
sLogical net comprising physical subnet
(initially ARPANET, Telenet, Phonenet).
Advanced network services
(initially mail, ﬁle transfer, remote
login, name server).
Self-governing, -sustaining, and -supporting
Low entry fee
DUES FOR CALENDAR 1983
Industry site -- $30,000
0Government site -- $10,00
Non-Proﬁt site -- $10,000
University site -- $5,000
ote: Sites with a very small number of researchers can negotiate lower dues.
Goals and Dues Structure of CSNET.
The four project teams must carry out these goals within two important constraints. First, the
total project budget, $5 million over ﬁve years, places a severe limit on the number of personnel wh
an be hired. It also means that satisfactory service must be available by 1983 to permit collection of
dues. Second, the project must develop its own stable management organization by Spring 1983, whe
SF withdraws ofﬁcially from project management.
Users who have accounts at ARPANET hosts already have the full services of CSNET except for
the name server. Until Telenet sites are operational, all other users are at Phonenet sites; they will not
ave ﬁle transfer or remote login services and will interact with the name server by mail. At the start
of the project, mailboxes are also provided on a machine called the CSNET Public Host for users who
ave no accounts at ARPANET or Phonenet sites; this service is not heavily used.
C -7-SNET Overview (12/1/82)
CSNET protocol and name-server software development is limited to the Berkeley UNIX operat-
ng system on VAX computers. The MMDF software works with Berkeley UNIX and Bell Labs’ Ver-
sion 7 UNIX. CSNET is encouraging vendors to develop compatible software and hardware for other
achines and operating systems.
Figure 2 lists the three technical subprojects of CSNET: Phonenet, Name Server, and Protocols.
The following subsections give overviews of these projects. Companion papers describe these project
n detail [1,2,3].
The Phonenet project is being conducted jointly at the University of Delaware and the Rand Cor-
poration. Its ﬁrst goal is to set up and operate mail relay computers on the east coast (at the University
f Delaware) and on the west coast (at the Rand Corporation). Its second goal is to provide each
CSNET site with the MMDF mail transport to permit automatic mail exchange between that site’
achine and the nearest relay.
Each relay will route mail to the destination site via ARPANET, Telenet, Phonenet, and possibly
the other relay. CSNET sites can poll the relays as often as they desire and are willing to pay tele-
hone line charges. Phonenet messages can incur usage charges if they are routed over paths for which
For example, the IBM Corporation has an agreement with the University of Wisconsin to develop CSNET-compatible
protocols for IBM machines running the VM operating system. Also, a Pascal version of the Phonenet software has
nabled Phonenet participation by users of DEC VMS and IBM VM operating systems.
)CSNET Overview (12/1/82-8-
PROJECT INVESTIGATORS GOAL
Phonenet D. Farber (Delaware)
A. Hearn (Rand)
Install and operate mail relays (VAX
11/750) at Delaware and Rand, connecte
y phone, Telenet, and ARPANET. Distri-
bute copies of MMDF software to CSNET
honenet sites. Goal is 100 operational
sites by 1983.
L. Landweber (Wisconsin)
M. Solomon (Wisconsin)
Develop the CSNET name server, a data-
base of all CSNET users, and install it on
he CSNET Service Host. Final version
distributed to CSNET sites by end of 1983
D. Comer (Purdue)
P. Denning (Purdue
. Korb (Purdue)
Construct interface between ARPA’s Inter-
net Protocol (IP) and the X.25 public net-
ork protocol to permit using X.25 nets
for full ARPANET services. Working ver
ion available for distribution by 1983.
Technical Subprojects of CSNET.
.CSNET must pay, e.g., telephone lines or Telenet
The design goals of MMDF have been reported at the 1979 Data Communications Symposium
. They are: 1) a mail transport that provides a high-level, channel-independent interface, 2) erro
hecking of message formats, and 3) robustness under load. The main components of MMDF are illus-
trated in Figure 3.
The ﬁrst goal is achieved by implementing the delivery mechanism as a process that takes mes-
sages one at a time from its work queue and sends them over one of several channels connected to it
utput. The incoming messages must consist of a body and a header in a standard (internal) format.
The delivery process uses the address in the header to select a channel for sending the message. Eac
hannel is a driver that sends the message in the protocol of a particular network, e.g., local delivery,
C -9-SNET Overview (12/1/82)
ARPANET, UUCP, or Phonenet.
The second goal is met by requiring that every user mail environment contain a certiﬁed "submit"
subprogram for interacting with the MMDF delivery process. When a user forms an address in the
eader of a message, the submit subprogram converts it to the standard internal form and checks with
the delivery process to verify its validity; if the address is invalid, the user is immediately notiﬁed.
hen the user completes the message body, the submit subprogram deposits the ready message with
validated address in the work queue of the delivery process.
Mail incoming on any channel is also placed in the work queue of the delivery process, which
will eventually deposit it in a local mailbox via the local delivery channel.
The third goal is met by storing the work queue of the deliver process on secondary storage,
which can allow it to become quite large without overloading the system.
CSNET soon encountered difﬁculties on account of an essential incompatibility between the
MMDF transport and the one already in the UNIX operating system at each of the Phonenet sites. Th
egular UNIX mailer programs do no address checking; instead, the delivery process parses and inter-
prets addresses of many formats. There is no simple way to convert these mailers to interact with th
MDF delivery process directly. Because of this, many Phonenet sites had to deal with two mail sys-
tems -- the one already in their UNIX operating systems and the new one provided by MMDF. This
as a major inconvenience that created many complaints.
To provide a temporary solution, the Purdue and Delaware contractors cooperated on mino
odiﬁcations of delivery processes of Berkeley UNIX and MMDF. The outgoing ARPANET channel
on the Berkeley delivery process was replaced with a driver that hands mail intended for the
RPANET to the MMDF delivery process, which in turn routes it to its destination via the Phonenet
channel. (Because Phonenet sites cannot send directly to the ARPANET, this loses no function.) Th
MDF local delivery channel was modiﬁed to store mail in the mailboxes and in the formats expected
eby the Berkeley mail programs. The disadvantage of this solution is that it requires each Phonenet sit
CSNET Overview (12/1/82
o install and maintain two mail transport systems.
A long-term solution is being discussed between CSNET and the Berkeley UNIX developers.
The goal is a single mail transport that incorporates the best features of Berkeley’s current mail syste
nd MMDF. This system would be distributed with releases of Berkeley UNIX.
Name Server 
The name server is a database of all CSNET users held online at a site called the CSNET Service
Host. (This machine is now at the University of Wisconsin.) Each record of this database contains the
ame, mailbox identiﬁer, and descriptive keywords of a registered CSNET user. CSNET sites can
query the name server to obtain the mailbox address of any user. CSNET users can update the key
ords in their records at any time. The name server project is also implementing the programs to be
installed at each CSNET site for the proper protocols with the name server.
The user interface with the name server consists of commands to implement these operations:
The "register" command is used by a user to enter a new record in the database; the "unregister" com
and is used to remove a record; the "move" command is used to change the ﬁeld in a record indicat-
ing the location of the user’s mailbox. A password scheme prevents unauthorized use of these com-
ands. The "whois" command is used to retrieve a set of records matching keywords; for example
whois Peter Denning
CSNET Overview (12/1/82
whois past ACM president
will return the same record. The mailbox identiﬁer ﬁeld of this record can be extracted and put in
ocal alias table so that future queries can be bypassed. The "update" operation is used by a user to
alter the descriptive keywords in his record of the database; the system requires him to present his logi
assword before installing any changes.
Future versions of the name server software distributed to CSNET sites will automatically
encache information locally to reduce trafﬁc with the name server. For example, a user’s alias tabl
ill hold pairs (nickname, mailbox-identiﬁer) to help avoid unneeded invocations of the whois com-
mand by that user. A system table will encache pairs (mailbox-identiﬁer, internet address) so that
nneeded requests for the internet address of a given site can be omitted. Messages will be automati-
cally forwarded to users whose mailboxes have been moved.
Users at Phonenet sites will have to conduct the above interactions by sending mail to the name
server. The name server will, by return mail, send records matching a whois query.
The IP-to-X.25 protocol project is the least visible component of CSNET. Its goal is an interface
between the datagram-oriented DARPA Internet Protocol (IP) and the virtual-circuit-oriented X.25 pub-
ic packet network protocol. This interface will enable the high-level ARPA Transport Control Protocol
(TCP) to work with X.25 nets, which in turn will allow CSNET users full access to ARPANET ser-
ices. Because all user services interact with TCP, any new DARPA network services will become
available throughout all of CSNET with new software distributions.
-12- )CSNET Overview (12/1/82
DARPA’s protocol design relies on the TCP layer in one machine to establish a process-to-
process channel with a corresponding TCP layer on another machine. The sender’s TCP breaks a mes
age into packets, which are handed over to its IP for transmission as independent datagrams over the
ARPANET. The receiver’s IP hands the received datagrams to its TCP, which reassembles them into
essages. Purdue’s modiﬁcation extends IP so that it can selects the Purdue X.25 interface when the
recipient host’s internet address is on Telenet.
Because Telenet charges for opening virtual circuits in X.25, the Purdue interface cannot transmit
an IP datagram simply by opening a circuit, sending a packet containing the datagram, and then closing
he circuit. Instead, it must leave open a circuit to the target host as long as it or the target is actively
using that circuit. An algorithm resembling a page replacement algorithm for a virtual memory syste
loses an open circuit when IP requires a new circuit beyond the maximum number Telenet permits a
given host to open.
The Purdue interface is connected to an Interactive Systems INcard, which is a board that con-
nects the Telenet modem to the VAX backplane. A future project is to provide the X.29 protocol
xtension of X.25; this will allow the INcard to connect to a login port of UNIX so that authorized
users can access the machine by ordinary Telenet remote login. Interactive Systems is considering
CSNET Overview (12/1/82
xtending the design of its INcard to include X.29 on the board.
Organization of the Project
The CSNET management structure provides central management over a project whose com-
ponents are at different locations. Figure 4 outlines the management components.
R. Adrion (NSF)
L. Landweber, Chair
C. W. Kern
)Policy (L. Landweber, Chair
Technical (D. Farber, Chair)
Organization (A. Hearn, Chair
oordination & Information Center (R. Edmiston, Director)
F CSNET Management Structure.IGURE 4:
The Management Committee consists of the principal investigator of each contract, the director of
Until October 1982, C. W. Kern of NSF was the project director.
)CSNET Overview (12/1/82-14-
the Coordination and Information Center (CIC), and the NSF project manager. This committee meet
very six to eight weeks to review the status of the project, settle policy questions, and give guidance to
the technical subprojects. It is responsible for keeping the entire project on schedule and for initiating
orrective action when needed. Until a separate CSNET organization is established in 1983, the NSF
project director can overrule the committee on any matter.
The Policy Support Group consists of the Management Committee plus other senior computer
It meets as needed (but at least once a year) to review the status of the entire project, giv
eneral guidance to the Management Committee, and determine overall policies for CSNET. When
CSNET becomes a separate corporation, this group will be replaced by a board of directors and the
anagement Committee will be replaced by an executive committee.
The Technical Support Group meets from time to time to consider technical problems facin
SNET and recommend solutions to the Management Committee.
It consists of the principal investi-
gators of each project (or their designees) plus other computer scientists whose technical areas are
elevant to the project.
The Organization Support Group has been studying models for the CSNET organization -- e.g.,
consortium of universities, embedding in an existing consortium, or separate corporation.
onstitution and bylaws that is serving as the charter of CSNET until a formal organization is esta-
blished. After considering the alternatives proposed by this committee, the NSF decided that embed
ing CSNET into an existing organization would be the best choice; it issued a program solicitation in
As of September 1982, the other members of the Policy Support Group were A. Aho (Bell), B. Arden (Princeton), J.
Birnbaum (HP), F. Corbato (MIT), K. Curtis (NSF), R. Eckhouse (DEC), N. Habermann (Carnegie-Mellon), R. Kahn
DARPA), L. Kleinrock (UCLA), M. Marcus (FCC), R. Miller (Georgia Tech), R. Ritchie (Washington), H. Schorr
(IBM), R. Spinrad (Xerox), and S. Sedelow (Kansas).
As of September 1982, the members of the Technical Support Group were E. Allman (Berkeley), V. Cerf (MCI), D.
Crocker (Delaware), P. Enslow (Georgia Tech), J. Feldman (Rochester), L. Hollaar (Utah), J. T. Korb (Purdue), K.
antz (Stanford), M. O’Brien (Rand), J. Postel (USC), L. Rowe (Berkeley), F. Schneider (Cornell), and M. Solomon
As of September 1982, the members of the Organization Support Group were A. Batson (Virginia), W. Franta (Min-
nesota), M. Harrison (Berkeley), G. Heller (EDUCOM), L. Travis (Wisconsin), and K. Uncapher (USC).
CSNET Overview (12/1/82)-15-
.October 1982 inviting potential host organizations to submit proposals
The Coordination and Information Center is to provide the management services needed to regis-
ter and install new users and sites, distribute CSNET software, provide documentation and regular
ewsletters, and answer users’ questions. After a public announcement in Fall 1981, NSF received and
reviewed proposals. A contract was awarded to Bolt Beranek and Newman of Cambridge, MA, for thi
unction. The CIC became operational in July 1982.
The CIC has been advising the Management Committee on possible methods of accounting for
CSNET use and billing sites their fair shares of these charges. The accounting problem is complicate
y several factors: a) Some messages will travel through several subnets each with different charging
policies. For example, an east coast Phonenet site will communicate with a west coast Phonenet site b
wo phone calls (site to Delaware relay, Rand relay to site) and a Telenet call (relay to relay). b) It is
impractical for the relays to provide itemized lists of load generated by each user at a site. It can pro
ide a total of load generated by a site. The site will have to allocate the cost of that load locally
among its users; CSNET will provide accounting software for this purpose. c) By agreement wit
ARPA, CSNET will not charge ARPANET users for CSNET use. Assuming that trafﬁc from CSNET
to the ARPANET is approximately the same as trafﬁc from ARPANET to CSNET, an approach would
e to charge CSNET users twice the cost of the CSNET leg of their messages to ARPANET users. d)
Duplex telephone and Telenet circuits can be used by a receiver to send mail back to a caller -- at the
aller’s expense. Special controls may be needed if some frequently called sites are found not to be
paying usage fees proportional to their actual outbound trafﬁc.
CSNET Overview (12/1/82
tatus of the Project
As of October 1982 both the Rand and Delaware relays were operational. The two relays com-
municated by telephone and ARPANET, and will communicate by Telenet as soon as practicable.
ome 76 Phonenet sites were operational or about to become so, as shown by the CSNET map in Fig-
As of September 1982, a preliminary version of the name server software was being tested at the
contractors’ sites. All users at all operational CSNET sites were registered in the database. All the
ommands in the user interface were implemented. Later versions making use of the alias facility and
of encaching internet addresses were under development but were not yet under test.
As of September 1982, the protocol software was under test and was being installed at each of
the other contractor’s machines. During the test period, all communication among the contractors wil
se Telenet instead of ARPANET. Around January 1983 this software will be available for other
CSNET sites to install if and when they obtain the necessary Telenet connections. Initial tests reveale
hat Telenet window restrictions limited effective throughput to less than 1200 baud even on lines rated
at 9600 baud. Negotiations with Telenet were initiated to loosen these restrictions.
As of July 1982 Bolt Beranek and Newman had been accepted under contract to provide the
Coordination and Information Center. Complete plans for documentation and software distribution wer
pproved by the Management Committee and were being implemented. A hotline phone number was
operational for any CSNET user or site having questions or comments (617-497-2777). The CIC per
onnel had assisted the management committee in devising a dues structure and were assisting in the
development of accounting and billing procedures.
As of September 1982 the Policy Support Group had approved a draft of the constitution and
bylaws of CSNET. Until CSNET is converted to a separate organization, this document will serve a
he charter. In October 1982, the NSF issued a formal solicitation for an institution to serve as host for
CSNET, Inc., during the next period.
-17- )CSNET Overview (12/1/82
As of September 1982 the Management Committee had received approval from NSF for the dues
structure (Figure 1) and was proceeding to its implementation. During 1983 member institutions will
egin paying dues and usage fees. The Coordination and Information Center has provided estimates of
the annual usage costs that can be expected by each type of site. Examples of costs from these esti-
ates are summarized in Figure 6.
Initial Annual Annual
quipment Connect Usage Annual
Site Type Cost Charges Charges COST
PHONENET 1500 250
Moderate (1) 8750 9000
Heavy (2) 24250 2450
ELENET 10000 12000
Moderate (1) 3250 1525
eavy (2) 9000 21000
ARPANET -- 107,000 -- --
(1) Moderate = 10 moderate users plus 10 heavy users.
2) Heavy = 20 moderate users plus 30 heavy users.
F Example Usage Cost Estimates.IGURE 6:
(These examples assume a moderate user will generate about $250/year charges at Phonenet sites and
$75/year at Telenet sites; they assume a heavy user will generate about $625/year at Phonenet sites an
250/year at Telenet sites.) Phonenet usage charges are a combination of telephone line charges and
Telenet packet charges for messages that have Telenet legs in their journeys. For a sufﬁciently activ
ite, Telenet is cheaper than Phonenet. Moreover, Telenet gives interactive access to CSNET services;
Phonenet does not. The ARPANET ﬁgures are included for comparison: CSNET will be able to pro-
ide similar services at much lower cost to many members of the computer research community.
-As of September 1982 the Management Committee had admitted to membership only U.S. institu
CSNET Overview (12/1/82
ions conducting or directly supporting computer research. Several industrial applications from ﬁrms
engaged in product development had been deferred until the CSNET organization is more stable and ne
se policies have been set forth. Several applications from foreign research sites (in Canada, Israel, and
Europe) had been deferred until the U.S. government’s policies on transborder ﬂow in respect to
SNET can be ascertained.
We are grateful to the many colleagues with whom we have interacted during the formative
stages of CSNET, to the people at NSF and DARPA for their constant advice and encouragement, t
he National Science Board for its faith in computer community, and to the computer community for its
continued support for this project. Three persons in government played critical key roles. Bob Kahn
nd Vinton Cerf of DARPA gave constant support and a major commitment of DARPA resources.
Kent Curtis of NSF spent uncounted hours behind the scenes selling the concept of CSNET to the NS
dministration, to OMB, and to DARPA; his insistence on wide community support signiﬁcantly
strengthened the project and assured its success. We are also grateful to Fred Schneider and Larr
andweber for comments on this manuscript.
. Comer, D. E., "The Computer Science Research Network CSNET -- A History and Status
Report," submitted to Communications of ACM, August 1982.
. Landweber, L., and Solomon, M., "Use of Multiple Networks in CSNET," Proc. IEEE COMP-
C -19-SNET Overview (12/1/82)
CON (February 1982).
. Crocker, D. H., Szurkowski, E. S., and Farber, D. J., "An Inter-Network Memo Distribution Capa-
bility -- MMDF," Proc. Sixth Data Communications Symposium, ACM SIGCOMM (1979), 18-
4. Landweber, L., Litzkow, L., Neuhengen, D., and Solomon, M., "Architecture of The CSNET
Name Server," Proc. Symposium on Data Communications, ACM SIGCOMM (March 1983), thi
5. Comer, D. E., and Korb, J. T., "CSNET Protocol Software: The IP-to-X.25 Interface," Proc. Sym-
posium on Data Communications, ACM SIGCOMM (March 1983), this issue.