Licklider’s Intergalactic Computer Network - The Computing Universe: A Journey through a Revolution - page 199

Baran's second idea was just as radical: to introduce a type of redundancy
in sending messages. He proposed splitting each message into a number of
fixed-length “message blocks” - what we now call packets - and allowing the
blocks to take independent paths through the network. In an interview with
the writer Stewart Brand, Baran later described what he called the “wonderful”
properties of this method, called packet switching:

Fig. 10.10. “Interactive Communication
consists of short spurts of dialog. F1 ”

Anyway, you send the packet out, and when this station gets it off to the next
guy, it sends back, “OK, got it. You can erase the previous one.” If the first
station doesn't hear back, it sends another copy out in a different direction.
The packets can arrive out of order. We just sort them out at the end. Since it
didn't have to be synchronous, you didn't have to lock everything all together.
It didn't take very long before we started seeing all sorts of wonderful
properties in this model. The network would learn where everybody was.
You could chop up the network and within half a second of real-world time it
would be routing traffic again. Then we had the realization that if there's an
overload in one place, traffic will move around it. So it's a lot more efficient
than conventional communications. If somebody tries to hog the network,
the traffic routes away from them. Packet switching had all these wonderful
properties that weren't invented - they were discovered. 9

In spite of all these advantages, Baran had a hard job trying to convince AT&T -
which then operated virtually the entire U.S. telephone network - of the vir-
tues of distributed networks and packet switching. At the time, all telephone
networks used what is known as circuit switching . When you called someone, the
network established a physical connection between the two endpoints and the
intermediate links were dedicated to your conversation as long as you stayed
on the line. For a phone call, this method makes sense, because there is typi-
cally a steady exchange of information during a conversation. By contrast, data
communications sent from a user sitting at a computer tend to be “bursty,”
with large amounts of data sent all at once, followed by pauses with no data
being transmitted ( Fig. 10.10 ). Keeping a line reserved for this type of communi-
cation is not a very efficient way to use the available bandwidth in the network.
Bandwidth is the technical term used to define a network's maximum capacity
for transmitting information. Using packets that can be routed in multiple ways
through a distributed network turns out to be a much more efficient use of the
available bandwidth because packets from different messages can use the same
links. Because the different packets of the same message can travel over multi-
ple different paths, the packets making up any given message may arrive out of
sequence and need to be reassembled in the right order at the receiving center.
Each packet must therefore contain a header at the beginning of the packet that
describes where it is going and to which message it belongs.
In response to criticisms of his ideas from his colleagues, Baran wrote a
series of papers that showed in detail how all these problems could be solved.
One new factor in understanding his idea was simply the much higher speed of
switching that was possible in computer networks. As we have seen, the elec-
tromagnetic relay switches of the earliest computers had been replaced by elec-
tronic switches capable of much-faster switching rates. To AT&T engineers not
familiar with the new digital computer technology, Baran's idea of breaking up