Imagine it is 1983. A computer science department wants to send a research question to a colleague on ARPANET. The department has a computer, a modem, and a telephone line. It has no direct ARPANET connection.
The computer can make a call, hand the message to a relay, and hang up. The relay takes it from there. A reply can wait at the relay until the department calls again.
That’s the idea behind PhoneNet, the dial-up email service of the Computer Science Network (CSNET). It brought researchers into the same conversation even when their institutions could afford very different kinds of connectivity. [1, 2]
CSNET History: Who Could Join the Network?
ARPANET had already shown what connected research teams could do. But its original purpose was to support ARPA-sponsored research, and many university computer science departments had no direct access. For those departments, the problem was not getting a faster connection. It was getting into the networked research community at all. [2, 3]
CSNET began operating in 1981 with support from the U.S. National Science Foundation (NSF). Larry Landweber, Peter Denning, David Farber, and Anthony Hearn led the effort, with support from NSF program officer Kent Curtis. It served computer science researchers at universities, companies, and government institutions. [3, 4]
The project faced a practical choice. Requiring every member to install the same kind of connection would have left some departments out. CSNET offered different levels of service instead. A department that needed remote login or file transfer could use an IP service. A department that mainly needed to exchange messages could start with PhoneNet. [1, 2, 5]
Follow One Message Through PhoneNet
Think of a relay as a post office that keeps mail for a town until its van comes to collect it. The town does not need a vehicle on the road all day. It needs somewhere to leave the letters and someone to make the next trip.
Here’s a simplified example based on the documented PhoneNet design. The department and the message are fictional.
- Write the message locally. A researcher submits email on the department’s computer. Mail software puts it in an outgoing queue. The telephone call does not have to happen while the researcher is typing.
- Call the relay. At a later poll, the department’s computer dials a CSNET relay. The call lets it hand over outgoing messages and collect mail the relay is holding for the site.
- Forward over the next available path. The relay accepts the message and can pass it into the mail system reachable via ARPANET. The department can disconnect after the handoff; the relay can finish the delivery without a live line back to campus.
- Wait for the next call. The colleague’s reply goes to the relay serving the department. The relay holds it until the department connects again, after which the site’s own mail system can deliver it locally. [1, 6]
This is store-and-forward. Each stage can hold a message until the next stage is ready. Notice who makes the call: the campus host typically polls the relay. The researcher does not have to sit at the telephone waiting for a reply.
The wait was a real trade-off. Suppose a site calls at 8 a.m. and noon. If a reply reaches its relay at 8:05 a.m., it could wait there three hours and 55 minutes for pickup. That’s an illustrative schedule, not a historical polling interval. Sites could choose when to poll, balancing the need for timely delivery against telephone charges. [1]
There is a small piece of historical evidence that makes the process tangible. A technical memorandum preserved by NASA reproduces a period help screen showing how a user could address mail through UDEL-RELAY or RAND-RELAY. The example ends with the status word queued. The system had accepted the message for forwarding; it was not claiming that someone at the other end had read it. [6]
PhoneNet Was One Way In
PhoneNet reached sites by exchanging messages. Those sites did not need a general-purpose IP path for another host to reach their mail users. That did not make their computers remotely accessible over PhoneNet for arbitrary applications.
Other CSNET members did want network-level connectivity. The project also developed a way to carry TCP/IP traffic across a commercial X.25 packet network, including GTE Telenet. And some CSNET participants already had access to ARPANET. The service could span those arrangements without asking every member to upgrade at once. [1, 2, 7]
| Example access arrangement | What it offered |
|---|---|
| PhoneNet dial-up | Periodic mail exchange with a relay. A site could send and receive messages without a general IP connection through PhoneNet. |
| IP over X.25 | IP connectivity over a commercial packet service. With the appropriate software and access, a site could use services such as remote login and file transfer. |
| Existing ARPANET connection | An already attached participant could use its ARPANET connection. Mail relays gave it a way to correspond with researchers at PhoneNet sites. |
A related idea appears in our account of the 1977 Three-Network Test: participating networks can differ internally and still exchange data. The difference here is the level at which the systems met. The 1977 experiment demonstrated internetworking. A PhoneNet site could exchange email through a relay without itself having the same network-layer service as the recipient.
The Relay Was Where the Systems Met
The project used MMDF, mail software that could queue a message and select a delivery channel based on where it was going. One channel could hand mail to a local mailbox; another could send it toward a remote system. At the University of Delaware, Dave Crocker worked on PhoneNet; at RAND, Michael O’Brien worked on the relays that connected CSNET and ARPANET. [1, 4]
That design kept the telephone call separate from the whole journey. Mail could cross an intermittent modem link on one leg and use a different transport on another. The message survived the handoff even though the connection that brought it to the relay ended.
The bridge also depended on permission. An agreement in 1981 enabled CSNET traffic to use ARPANET infrastructure. That mattered as much as having a relay that worked: it made shared use an operational arrangement, not merely a technical possibility. [8]
Why a Mail-Only Service Still Had Users in 1989
At the January 1989 IETF meeting, Craig Partridge of BBN reported 183 CSNET members: 78 using PhoneNet and 105 using one of the IP services. The same report says the project had become financially self-supporting by 1985. [5]
Those figures are a snapshot of membership, not a count of personal accounts or modems. They show something easy to miss in a history organized around ever-faster links: years after CSNET began, more than two in five members in that snapshot still used its mail-only option. Fuller connectivity and periodic email served different needs.
CSNET and NSFNET were separate projects. NSFNET began operating in 1986 as a backbone linking NSF supercomputer centers and grew to connect regional and academic networks. CSNET had already demonstrated the demand for broader research networking and shown that organizations could participate at different levels. Our history of networking in the 1980s puts both projects in the wider chronology. [3, 4]
The ARPANET story explains how researchers learned to share resources across a network. CSNET faced the next question: who else could join that conversation, and what was the smallest useful connection they could start with?
For some universities, the answer was a phone call. The call could end, and the message could keep going.
References and Further Reading
- Peter J. Denning, Anthony Hearn, and C. William Kern, History and Overview of CSNET — December 1982 technical report and preprint of the 1983 paper; describes the PhoneNet relays, polling, MMDF, and IP over X.25.
- Charles Babbage Institute: An Interview with Peter J. Denning — a participant’s account of the service tiers and the project’s design.
- National Science Foundation: The Internet — NSF’s account of its support for CSNET and the launch of NSFNET.
- Internet Society: CSNET Receives the 2009 Jonathan B. Postel Service Award — project leadership, institutional roles, and legacy.
- Craig Partridge, CSNET Report, January 1989 IETF Meeting — a contemporary update on operations and membership by connection type.
- Stuttgart Interconnection Network Project from PIX to NICS, NASA Technical Memorandum 77802 — Figure 7 reproduces the historical
HELP CSNETscreen and relay-addressing example. - FCCSET Report to Congress, Internet Concepts, June 1987 — section 3.1.2.1 describes CSNET’s service levels and its use of commercial X.25.
- Barry M. Leiner et al., A Brief History of the Internet — a participant-authored history describing the 1981 agreement to share ARPANET infrastructure.
