Paul Baran · RAND, 1964

II

In 1960 the system with which the United States commanded its nuclear forces was high frequency and long distance, and it passed through a few switching centres.

The calculation that worried RAND was not how many cities would survive a first Soviet strike. It was another, narrower and more uncomfortable: if the Soviet Union destroyed those centres, the president would not be able to give the order to respond.

From there comes the whole problem. A retaliation that cannot be ordered does not deter, and a deterrence that does not work invites precisely the attack it meant to prevent. The command system was the point at which the whole apparatus became useless, and it was also the easiest to reach.

Paul Baran worked in RAND's mathematics division and spent two years on it. What he delivered was not a plea about the freedom of communications but a survivability study, and the question was how much network has to be destroyed for an order to stop being transmissible.

III

The technical part of the finding is more interesting than the drawing; it is also the part almost never cited.

Baran defined a number he called the level of redundancy: how many links each station has, on average, above the minimum needed for the network to be connected. Level one is a tree, the network that breaks entirely if any branch is cut. Level eight would be an expensive and absurd mesh.

Then he simulated attacks. He destroyed stations at random, destroyed links, destroyed both, and measured what fraction of the surviving nodes could still talk to what fraction.

The result is the one that turned the report into something that had to be read. With a level of redundancy of three or four —that is, with each station connected to three or four neighbours, which is few and is cheap— the network withstands the destruction of half its nodes, and the survivors remain in communication, almost all with almost all.

A perfect mesh was not needed. Very little mesh was needed, and that is the fact that holds up the right-hand panel.

On top of that topology Baran put the rest. Splitting each message into blocks of one thousand and twenty-four bits, sending each block on its own, letting each node decide where to pass it on according to a table that the traffic itself kept correcting. A node that received a block and did not know what to do with it sent it anywhere and the whole thing worked just the same. He called it hot potato routing.

What he was describing, under another name, is packet switching.

IV

Baran proposed building it. The air force backed the project and commissioned an independent review, which came out favourable.

The execution, however, fell not to them but to the Defense Communications Agency, which at that moment had no experience in anything digital. Baran calculated that building it badly was worse than not building it, because a failure would bury the idea for a decade; he withdrew the proposal himself.

Before that he had been to see AT&T, which was the one that had the cables.

The telephone company's answer has been told many times and it is best not to embellish it. Its engineers held that the thing could not work, that a network where messages are split and each piece looks for its own way is not a communications network, and that establishing a circuit and holding it was the only serious way of doing things. Baran recalled that one of them told him he did not understand how a telephone worked.

It was not obstructionism and it was not blindness. It was a whole business —a century of circuit switching and of charging by the minute of established call— looking at a proposal that made it unnecessary and concluding what is usually concluded in that case.

V

Five years later, in 1969, the ARPANET was switched on, using packet switching.

Here two things that have been stuck together for half a century have to be separated. The ARPANET was not built to survive a nuclear war: it was built because research computers were extremely expensive and scarce, and the aim was for several groups to be able to use the same ones. Those who made it have repeated this many times and they are right.

What does come from the bomb is the technique. Baran had designed the splitting of messages and distributed routing so that a retaliation order could make its way through a half-destroyed network, and that technique turned out to be also the cheap way of sharing an expensive machine. Donald Davies, at the British National Physical Laboratory [Teddington], had arrived at the same thing on his own and without thinking about any bomb, and gave it the name that stuck.

So the correct sentence is neither that the internet was made to withstand a nuclear attack, nor that nuclear survivability had nothing to do with it. It is that the method was invented for that and adopted for something else.

VI

The drawing remains.

What was built after 1969 was not the right-hand panel. Intercontinental traffic concentrated in a few cables. Routing was left in the hands of a dozen networks that exchange the world between them without charging one another, the names of the world hang from thirteen letters [A to M] shared among twelve operators and most of what is served passes through a handful of companies that put their machines out in front.

Each point on that list is an economic decision taken separately, and none was a conspiracy. Concentrating is cheaper than distributing, always and at every step, and nobody has to want a centre for one to appear.

Baran died in 2011. By then his figure had spent decades illustrating texts about horizontal organisation, peer-to-peer networks and the end of hierarchies, in which the right-hand panel was presented as the description of something that already existed.

The right-hand panel is the only one of the three that nobody built. What there is is the middle one.