[e2e] Time for a new Internet Protocol
David P. Reed
dpreed at reed.com
Tue May 15 07:57:24 PDT 2007
A motivation for TCP and then IP, TCP/IP, UDP/IP, RTP/IP, etc. was that
network vendors had too much control over what could happen inside their
Thus, IP was the first "overlay network" designed from scratch to bring
heterogeneous networks into a common, world-wide "network of networks"
(term invented by Licklider and Taylor in their prescient paper, The
Computer as a Communications Device). By creating universal
connectivity, with such properties as allowing multitudinous connections
simultaneously between a node and its peers, an extensible user-layer
naming system called DNS, and an ability to invent new end-to-end
protocols, gradually a new ecology of computer mediated communications
evolved, including the WWW (dependent on the ability to make 100 "calls"
within a few milliseconds to a variety of hosts), email (dependent on
the ability to deploy end-system server applications without having to
ask the "operator" for permission for a special 800 number that
facilitates public addressability).
Through a series of tragic events (including the dominance of
routerheads* in the network community) the Internet is gradually being
taken back into the control of providers who view their goal as limiting
what end users can do, based on the theory that any application not
invented by the pipe and switch owners is a waste of resources. They
argue that "optimality" of the network is required, and that any new
application implemented at the edges threatens the security and
performance they pretend to provide to users.
Therefore, it is time to do what is possible: construct a new overlay
network that exploits the IP network just as the IP network exploited
its predecessors the ARPANET and ATT's longhaul dedicated links and new
technologies such as LANs.
I call for others to join me in constructing the next Internet, not as
an extension of the current Internet, because that Internet is corrupted
by people who do not value innovation, connectivity, and the ability to
absorb new ideas from the user community.
The current IP layer Internet can then be left to be "optimized" by
those who think that 100G connections should drive the end user
functionality. We can exploit the Internet of today as an "autonomous
system" just as we built a layer on top of Ethernet and a layer on top
of the ARPANET to interconnect those.
To save argument, I am not arguing that the IP layer could not evolve.
I am arguing that the current research community and industry community
that support the IP layer *will not* allow it to evolve.
But that need not matter. If necessary, we can do this inefficiently,
creating a new class of routers that sit at the edge of the IP network
and sit in end user sites. We can encrypt the traffic, so that the IP
monopoly (analogous to the ATT monopoly) cannot tell what our layer is
doing, and we can use protocols that are more aggressively defensive
since the IP layer has indeed gotten very aggressive in blocking traffic
and attempting to prevent user-to-user connectivity.
Aggressive defense is costly - you need to send more packets when the
layer below you is trying to block your packets. But DARPA would be a
useful funder, because the technology we develop will support DARPA's
efforts to develop networking technologies that work in a net-centric
world, where US forces partner with temporary partners who may provide
connectivity today, but should not be trusted too much.
One model is TOR, another is Joost. Both of these services overlay
rich functions on top of the Internet, while integrating servers and
clients into a full Internet on top of today's Internets.
* routerheads are the modern equivalent of the old "bellheads". The
problem with bellheads was that they believed that the right way to
build a communications system was to put all functions into the network
layer, and have that layer controlled by a single monopoly, in order to
"optimize" the system. Such an approach reminds one of the argument for
the corporate state a la Mussolini: the trains run on time. Today's
routerheads believe that the Internet is created by the fibers and
pipes, rather than being an end-to-end set of agreements that can layer
on top of any underlying mechanism. Typically they work for backbone
ISPs or Router manufacturers as engineers, or in academic circles they
focus on running hotrod competitions for the fastest file transfer
between two points on the earth (carefully lining up fiber and switches
between specially tuned endpoints), or worse, running NS2 simulations
that demonstrate that it is possible to stand on one's head while
singing the National Anthem to get another publication in some
More information about the end2end-interest