I’m reading Janet Abbate’s Inventing the Internet, an excellent history of the Internet starting with its origins as the ARPANET in the 1960s. The most interesting things I’ve learned about so far is the heated battled between the TCP/IP protocol, which was favored primarily by the computer science research community, and the competing X.25 protocol, which was favored by the telecom industry. Embarrassingly, I didn’t know anything about this argument before I picked up Abbate’s book. What’s striking about it is how similar it sounds to arguments today. From page 161:
The operators of public data networks argued that ARPA’s TCP/IP failed to provide adequate control over network operations. For instance, a Telenet spokesman noted that, whereas X.25 was capable of controlling the flow of packets from each individual connection, TCP could only act on an entire host’s output at once. If one of the network connections from a host malfunctioned and flooded a TCP/IP network with packets, the network’s only defense would be to cut off the entire host, thus unfairly penalizing the others users on that host. Users of the research network might accept the inconvenience with resignation, but paying customers of a public data network would certainly protest. With regard to the business of running a network, the [Post, Telephone, and Telegraph Authorities] pointed out that IP had not been designed to allow networks to exchange the type of information that would be required for access control or cost accounting… TCP/IP had not been designed for a network serving as a public utility, with service guarantees and access charges. X.25 had been.
The choice between X.25’s virtual circuits and TCP/IP’s datagrams was not simply a technical matter; it also shifted the distribution of control and accountability between public network providers and private computers owners. For the PTTs, virtual circuits meant they could fuarntee their customers better service and boost their own profits. Some of their customers were glad to be offered reliable data communications service with little effort required on their part. For more expert computer owners, however, virtual circuits raised the cost of network service and interfered with their ability to control their own data communications activities.
Needless to say, I’m not posting this over an X.25 network. I think the advocates of non-neutral networking should be worried about the parallels here. While there were doubtless many factors that contributed to TCP/IP’s near-total in the 1990s, the flexibility, decentralization, and low cost of TCP/IP networks was clearly a decisive factor. The arguments that some proponents of a tiered Internet make today are eerily similar to the pro-X.25 arguments Abbate describes above. While we shouldn’t rule out experimentation with smarter networks at the margin, it strikes me as highly unlikely that a centralized, telco-controlled, “smart” network would work any better today than it did in 1987.
And just to avoid misunderstanding: the technical question of whether a non-neutral network is a good idea is separate from the policy question of whether mandating neutral networks is a good idea. In my opinion, neutral networks are a good idea, but laws mandating them aren’t so great.