screw up the program. An expensive failure would make it difficult for a
more competent agency to later undertake the project."[3-28] Thus, this archi-
tecture of control -- centralizing innovation and protecting an existing
model of doing business -- would not be questioned by Baran's work. At
least not then.

///\\\

The internet is not the telephone network. It is a network of networks
that sometimes run on the telephone lines. These networks and the wires
that link them are privately owned, like the wires of the old AT&T. Yet at the
core of this network is a different principle from the principle that guided
AT&T. Like the principle Baran confronted, this principle affects what is al-
lowed and what is not. And like the principle that Baran confronted, this
principle has an effect on innovation.

First described by network architects Jerome Saltzer, David Clark, and
David P. Reed in 1981, this principle -- called the "end-to-end argument"
(e2e) -- guides network designers in developing protocols and applications
for the network.[3-29] End-to-end says to keep intelligence in a network at the
ends, or in the applications, leaving the network itself to be relatively
simple.

There are many principles in the Internet's design. This one is key. But it
will take some explaining to show why.

Network designers commonly distinguish computers at the "end" or
"edge" of a network from computers within that network. The computers at
the end of a network are the machines you use to access the network. (The
machine you use to dial into the Internet, or your cell phone connecting to
a wireless Web, is a computer at the edge of the network.) The computers
"within" the network are the machines that establish the links to other
computers -- and thereby form the network itself. (The machines run by
your Internet service provider, for example, could be computers within the
network.)

The end-to-end argument says that rather than locating intelligence
within the network, intelligence should be placed at the ends: compu-
ters within the network should perform only very simple functions that are
needed by lots of different applications, while functions that are needed by
only some applications should be performed at the edge. Thus, complexity
and intelligence in the network are pushed away from the network itself.
Simple networks, smart applications. As a recent National Research Coun-
cil (NRC) report describes it:

[[34]]

p033 _ -chap- _ toc-1 _ p034w _ toc-2 _ +chap+ _ p035

v?