Networks and Application Innovation

One of the truisms of modern Internet policy holds that innovation takes place primarily – or even exclusively – at the network edge. This idea, buttressed by appeals to the “end-to-end principle,” insists that technical improvements below the Internet Protocol layer are relatively insignificant except as they make the user experience faster. Consequently, there’s a strong element in the policy discourse toward bending over backward to make application innovations easier and network innovations harder. The idea of Internet exceptionalism holds that the Internet is the greatest boon for innovation ever because it passively moves information between edge-connected users and edge-connected service providers without doing anything important internally.

Like most popular ideas, this one doesn’t face much criticism and is generally accepted as gospel: innovators build applications, not networks. This idea isn’t entirely true, however. Consider the case of e-mail, one of the most advanced and wonderful applications the Internet has brought us. The story of e-mail’s advent is fairly well known:  a fellow named Ray Tomlinson combined a program called SNDMSG with program called CPYNET and voila, e-mail was born. SNDMSG allowed users of a single computer (running the TENEX operating system) to leave messages for other users of the same computer, and CPYNET copied a file between connected computers. Tomlinson’s innovation quickly became the most popular application on the ARPANET, and versions of electronic mail were quickly developed for other operating systems and networks.

There’s no question that e-mail was an important application that spurred the adoption of ARPANET. It’s not an exaggeration to say it was the first “killer app” for networked computing. Here was an application that allowed people to communicate with each other for virtually no cost (ignoring the computers and networks involved, of course) irrespective of time zones and independent of the other restrictions that the telephone involved, such as its tendency to ring when you least want to be interrupted. But was this really new?

E-mail is strikingly similar to the telegram. In the early days, e-mail was limited to plain text – graphics and formatting were introduced much later (and a bit painfully) – just as the telegram was. They tended to be short and to the point, and while they were delivered quickly, there was no expectation of an immediate reply.

Teletype ASR-33Oddly, there’s a direct continuum between e-mail and the telegram at the network edge that we rarely think about. The terminal in the picture is a Teletype Model ASR-33 teleprinter. This terminal was more or less the standard terminal for the time-sharing computers of the 1960s and 70s for which ARPANET was developed, so if you sent somebody an e-mail across ARPANET, it was most likely typed on a Teletype at the sending end and printed on a Teletype at the receiving end, all in uppercase characters, with limited punctuation and no graphics.

Before, during, and even after the reign of ARPANET, there was another system of communication in use between businesses that involved a very similar interaction: people typed messages that were remotely printed across a network from one Teletype terminal to another. The businesses didn’t need to use ARPANET or to employ compatible programs on their computers as these messages moved across an entirely different network called Telex.

The Telex network was provided in the United States by Western Union, using many of the same facilities WU had built for telegraphy. The innovation behind Telex was the ability for customers to send messages directly to other customers without the last mile delay of the classic telegram – a delivery boy on a bicycle – and without mastering the Morse Code key. The Telex network could do this because it incorporated a message switch that automatically routed messages from the originator through the Telex Exchange to the recipient in real time; it was a hybrid of the all-digital telegraph network and the switched (analog) telephone system. And before Telex, the telegraph networks used Teletypes to send and receive telegrams from one Western Union office to another. The Teletype was a replacement, circa 1920, for the telegraph keys, pads of paper, and pencils that were the original “terminals” for the telegraph network. (note: When I started writing data communication software in the mid-1970s, one of my first tasks was to make a desktop computer emulate the ASR-33; connecting to packet-switched networks such as Telenet and Tymnet were among the bonuses of doing this, and it was also quieter to “print” letters on a video display than to print them on paper.)

We regard the functional difference between the telegram and the e-mail as so major that we consider e-mail a major innovation in the way the world communicates, but the innovation didn’t take place at the network edge. The telegraph operator sending a telegram for a Western Union customer in 1920 interacted with his Teletype terminal in substantially the same way that the users of ARPANET did when they composed messages and sent them to their peers in 1975.

The differences between the telegram, the Telex, and the e-mail were all in the networks that connected the Teletypes with each other. In the telegraph era, the networks were all private lines owned by Western Union (and similar companies in other countries) and there was no automatic switching. If a destination was particularly popular, as the stock exchange was, it got a WU office built nearby, but nobody was allowed to touch the WU network but WU. Telex relaxed these restrictions and allowed anyone to communicate directly with anyone, provided they were all Telex customers with compatible terminals, and the Internet relaxed the restrictions a bit more by allowing anyone on a compatible network (of which there were only three in the early days) to communicate regardless of terminal.

So the advance in interpersonal communication that is e-mail was an advance in network design more than it was an innovation at the network edge. When we send e-mail today we’re using an application that was devised in the mid-19th century and has been improved in substantial ways only as networks have improved. The increased level of utility from the 19th century telegram and the 21st century Tweet is mainly brought to you by network innovation, enhancement, and evolution.

You might want to bear that in mind the next time you’re encouraged to consider applications innovative and networks regressive.

  • Seth Finkelstein

    Well … yes and no … I won’t say you’re attacking a strawman, but I do think you’re spending much ammunition shooting down a weak argument.

    The real “edge” argument, in my view, is that application developers should *generally* not have to ask permission from incumbent network operators, as such incumbents have very anti-competitive incentives and anti-innovation bureaucracies (the manager can’t lose for saying “no”, but can be punished for saying “yes”). Obviously, there are problems in the other direction too, e.g. BitTorrent/P2P. But, on the whole, I think you give too little consideration to the value of strong nondiscrimination and open-access requirements.

  • Kaylie Volkman

    Thanks for the article post.Much thanks again. Awesome.

Comments are closed.