Surveillance or Security?: The Risks Posed by New Wiretapping Technologies (5 page)

Initially networks were quite local. The subscriber would ring the local
switch and tell the operator the name of the party with whom they wanted
to speak. The first switches were manual, consisting of panels with jacks
and cables between them. The operator would ring that party and then
connect the two lines on the switchboard via patch cords. While the original operators were teenage boys, their antics soon made clear that more
responsible people were needed, and young women became the telephone
operators of choice.' A Missouri undertaker designed the first automated
telephone switch."

We tend to think of a phone number as the name of the phone at
a particular location, but it is actually something else entirely. As Van
Jacobson, one of the early designers of Internet protocols, once put it, "A
phone number is not the name of your mom's phone; it's a program for
the end-office switch fabric to build a path to the destination line card.i11

Figure 2.1 (opposite page)

Centralized versus decentralized networks. Illustration by Nancy Snyder.

Consider, for example, the U.S. telephone number: 212-930-0800. The
first three digits-the area code-establish the general area of the phone
number; in this case it is New York City. The next three digits, normally called the telephone exchange, represent a smaller geographic area.12 In
our example the last four digits are, indeed, the local exchange's name for
the phone. Taken as a whole, the set of ten digits constitute a route description; the switching equipment within the network interprets that information much like a program and uses it to form a connection.

The first thing a modern telephone-and I will start by describing just
landline phones-must do is signal that it is "off hook" and thus ready to
make a call. This happens when the receiver is lifted, which closes a circuit,
creating a dial tone and signaling the central office (the local phone
exchange). Then the subscriber can dial the phone number she wishes to
reach ("dial," of course, being an anachronism from the era of rotary telephones). When the central office receives this number, its job is to determine where to route the call.

If the call is local-that is, within the same area code-then the switches
at the central office need to determine which trunk line, or communication
channel, should be used to route the call to an appropriate intermediate
telephone exchange. This new exchange repeats the process, but this time
connects to the recipient's local exchange. Since the first three digits
denote the local exchange and are thus unnecessary, only the last four
digits of the number are transmitted. The local exchange determines if the
recipient's line is free; if so, it "rings" the line. If the recipient answers, her
receiver closes a circuit to the local exchange, which establishes the call.13
The speakers have a fixed circuit for the call, the one that was created
during the call setup.

This is, of course, a simplified example: the call did not use an area code,
let alone an international code. The other simplification is that the call
described above had only two "hops"-that is, it only went through two
telephone exchanges.

The key goal of the network design was to provide quality of voice
service. Engineers needed to factor in that each time a call goes through
an exchange, it needs to use a repeater to amplify the voice signal. Passing
through a repeater causes the signal to change slightly. Thus the network
needed to minimize the number of times a call would go through an
exchange. The telephone company limits calls to five hops, after which it
deems the degradation in voice quality unacceptable.14 Digital signals do
not face this problem and thus can travel through an arbitrary number of
repeaters. This small engineering difference leads to a remarkable freedom
in system design. Messages can traverse an arbitrarily long path15 to reach
a destination, enabling a more robust network.

The telephone system is built from highly reliable components. The
telephone company believed in service that allowed a user's calls to go through ninety-nine times out of a hundred. Since central office switches
served ten thousand lines, this meant "five 9s" reliability (otherwise the 1
percent blocking could not be satisfied). Of course, more than central office
switches are needed to service a call that travels between two destinations
with different central offices.

At the height of the Cold War, some engineers began thinking about
reliability differently. After all, you might care less about talking to a particular person at a particular moment than about getting the message
through eventually. That is, presuming the other party is in and willing to
answer the phone, you might not be concerned about always being able
to connect each time you dialed, but you might want to ensure that the
message you are attempting to send eventually gets through. This was the
problem that the designers of the Internet tried to solve.

2.2 Creating the Internet

In the 1960s, physicists had realized that the electromagnetic pulse from
a high-altitude nuclear explosion would disrupt, and quite possibly destroy,
electrical systems in a large area. Any centralized communication network
such as the phone system that the United States (and the rest of the world)
used would be in trouble." RAND researcher Paul Baran went to work on
this problem.

Since the frequency of AM radio stations would not be disrupted by the
blast, Baran realized the stations could be used to relay messages. He implemented this using a dozen radio stations." Meanwhile, using digital networks, Donald Davies of the United Kingdom's National Physical Laboratory
found another solution to the problem. Davies solved Baran's problem
while trying to address a completely different question.

Davies was interested in transmitting large data files across networked
computers.18 The problem is different from voice communications. Data
traffic is bursty: lots of data for a short time, then nothing, then lots again.
Dedicating a telephone circuit to a data transfer did not make a lot of sense;
the line would just not be used to its full extent19 and, unlike with voice,
a small delay is not a major issue in transmitting data files.

Both Baran and Davies hit upon the same solution. Redundancy in the
network-multiple distinct ways of going between the sender and the
recipient-was key. Such redundancy is surprisingly cheap to obtain. Say a
network has redundancy 1 if there are exactly enough wires connecting the
nodes so that there is one path between any two nodes; if there are twice
as many wires, call that redundancy 2, and so on.20 Trying experiments
on more traditional communications networks, Baran ran simulations and discovered that with a redundancy level of about 3, "The enemy could
destroy 50, 60, 70% of the targets or more and [the network] would still
work.i21 The design resulted in a highly robust system.

The distributed networks that Davies and Baran had independently
invented were to be even more decentralized than decentralized networks
of earlier efforts. Network redundancy meant that paths might be much
longer than the typical PSTN communication. Thus the communications
signal had to be digital, not analog. That turned out to be a tremendous
advantage. There was no need for the entire data transfer to occur in one
large message; indeed, efficiency and reliability argued that the message
should be split into small packets.

The idea was that when the packets were received the recipient's
machine sent a message back to the sender saying, "OK; got it." If there
was no acknowledgment, after a short period, the sender's machine would
resend the packet. Of course, because the packets traveled by varied routes,
they might arrive out of order. But the packets could be numbered, and
the receiving end could simply sort them back into order.

One of the striking things about this proposed network was that while
the network itself was to be extremely reliable, individual components
need not achieve that same level of reliability. Instead the network
depended on "structural reliability, rather than component reliability.""
Small amounts of redundancy led to vastly increased reliability, a result
surprising to the engineers.23

The Internet's decentralized control meant all machines on the network
were, more or less, peers. No one computer was in charge; the machines
were more or less equal and more or less capable of doing any of the
communication tasks. A computer could be the initiator or recipient of
a communication, or could simply pass a message through to a different
machine. This is the essence of a peer-to-peer network, and very much the
antithesis of the telephone company's hierarchical model of network
communication.

In Britain, the telecommunications establishment supported Davies,24
but in the United States Baran received a chilly reception from AT&T.
Baran was turning all the ideas that AT&T had used to manage their system
upside down. While scientists at the research arm of AT&T were quite
excited by Baran's work, corporate headquarters viewed that approbation
as the reaction of head-in-the-cloud scientists and refused to have anything
to do with Baran's packet-switched network.25 The odd thing about all this
was that the new network was not actually a new network at all. Baran
had built his network on top of the existing telephone network built by Alexander Graham Bell and his successors. It "hooked itself together as a
mesh,i26 simply connecting everything in new ways.

Scientific and technological ideas often emerge when the time is ripe,
and Baran and Davies were not the only ones to be considering packetswitched networks. (The actual term packet is due to Davies, who wanted
to convey the idea of a small package.) In 1961, Leonard Kleinrock, then
a graduate student at MIT, published the first of a series of papers analyzing
the mathematical behavior of messages traveling on one-way links in a
network. This analysis was critical for building a large-scale packet-switched
network.

One could say, only partially tongue in cheek, that the Internet is due
to Sputnik, the Soviet satellite that in 1957 startled the United States out
of its scientific complacency. In response the U.S. government founded the
Defense Advanced Research Projects Agency (DARPA), a Department of
Defense agency devoted to developing advanced technology for military
use.27 The Internet grew out of an ARPANET project and was perhaps the
most important civilian application that came from DARPA.

In 1966 DARPA hired MIT's Lawrence Roberts to build a network of
different computers that all communicated with one another.28 This would
be a resource-sharing network. Each individual system would follow its
own design with the only requirement being that the various networks be
able to "internetwork" with one another29 through the use of Interface
Message Processors (IMPs). Designing the IMPs fell to a Cambridge, Massachusetts, consulting company, Bolt Beranak and Newman (BBN), one of
whose researchers, Robert Kahn, moved to DARPA.

Kahn realized that only the IMPs would need a common language to
communicate, and this greatly simplified the entire scheme. The other
machines within the individual networks would not need to be transformed in any way in order to communicate with the rest of the system.
These principles made so much sense that forty years later they still govern
the Internet:

• Each individual network would stand on its own and would not need
internal changes in order to connect to the internetwork.

• Communications were on a "best-effort" basis. If a communication did
not go through, it would be retransmitted.

• The IMPs would connect the networks. These gateway machines (now
known as routers and switches) did not store information about the packets
that flowed through them, but simply directed the packets.

• All control would be local; there would be no centralized authority directing traffic.30 (This model is essentially opposite to the PSTN architecture of the time; there are many reasons for this difference, not all of them
technical.31)

With his colleague Vinton Cerf, Kahn developed the fundamental design
principles and communication protocols for the network that became the
Internet; they are the true "fathers of the Internet."

Kahn and Cerf developed the underlying protocol for end-to-end transmissions in 1973. It consisted of the Transmission Control Protocol (TCP)
and the Internet Protocol (IP) and is usually simply abbreviated as TCP/IP.
TCP determines whether a packet has reached its destination, while the IP
address is a numeric address that locates a device on a computer network.
Kahn and Cerf 's original version of IP used 32 bits for the address; this is
still in use today. As more and more devices connect to the Internet, the
32-bit address space, which allows for over four billion individual devices
to be located on the Internet, is running out of room.32 The new version
of addressing, IPv6 (Internet Protocol version 6), has 128 bits for addressing, which increases the number of possible Internet addresses by a factor
of 296, or more than a billion billion billion times as many.