The stupid network: Essential yet unattainable


                           Andrew Odlyzko
                        AT&T Labs - Research
                        amo@research.att.com
                   http://www.research.att.com/~amo

                         September 15, 1999


The stupid network is a myth.  Yet it is a powerful and useful myth
that should be the Holy Grail of networking.

If the current Internet were the stupid network, and provided just
dumb transport, then access to it would be a commodity.  However, that
is not consistent with the observed large price differences among
carriers.  The endless glitches in our communications also would not
be present in the stupid network.

While the stupid network is a myth, it is a useful myth with a large
grain of truth.  The Internet does allow users and developers to
ignore most of the complexities of the network.  By separating
applications from transport, it encourages the two to evolve
independently.  There is a short-term cost in the separation, since
the entire system is not as efficient at any time as it would be if it
were designed as a whole.  However, there is a huge long-term benefit
in flexibility.

The challenge for the future is to preserve the adaptability of the
Internet as it faces three related challenges.  The first will be
increasing mobility, as gadgets connected to the Internet increasingly
depart from the model of a PC tethered to the desktop.  The second
challenge will be posed by the rapid growth in devices connected to
the Internet.  The 40 microprocessors in the typical North American
household will be joined by many others, and will be increasingly
intelligent and acquire communication capabilities.

Finally, today's Internet provides largely simple point-to-point data
transmission.  The third challenge will come from the increasing
complexity of transactions that will be carried out online.  The
simple current operations, usually initiated by direct human action,
will be supplemented by interactions among a variety of partially
autonomous devices.  It will be the performance of entire systems, not
just communications links, that will have to be optimized.

To meet all these challenges, the Internet should strive to appear
even simpler than it is today.  Making it possible for our laptop to
communicate with the home security system and the plumber's PDA to
allow repairs to our house is an inherently difficult task.  It is
even harder than the task software producers face in getting different
packages to cooperate inside the same PC, and they certainly have
struggled with that chore.  Overengineering, as opposed to complicated
schemes that optimize network efficiency, is likely to win out, even
at the cost of efficiency.

While the Internet should appear as a simple network, it will need
sophisticated technical controls (such as the TCP congestion mechanism
that keeps the Internet from collapsing today) as well as the right
economic incentives.  Thus it will require much intelligence inside,
just as today's game consoles, although they appear simple to the
user, often have more computing power inside than the Cray-1
supercomputer.

The future of the Internet will be a competition between simplicity
and novelty, and while simplicity will be essential to enable novelty,
it is never likely to win completely.  The blame for this belongs to
us, the users, as we allow our requirements to grow.