[e2e] Cannara's views
David P. Reed
dpreed at reed.com
Fri Apr 13 08:27:00 PDT 2001
I changed the subject line because this has nothing to do with Outer Space.
For the students and others that might actually find Cannara's views
plausible, let me offer another clear point of view.
There are two things that Cannara says over and over.
1) The current internet protocol doesn't "optimize" the use of new
technologies (or incorporate those theoretical results that are closely
tied to the special properties of those cool new things).
2) New phenomena weren't anticipated in the original design.
(1) is true. The current internet protocol NEVER "optimized" the use of
any technology. So what? That was part of the design philosophy. In
order to be flexible enough to incorporate heterogeneity and change, you
can't optimize for a particular technology. More importantly, if you do,
you die at the next change.
(2) is wholly false. The original design philosophy was based on a
principle of humility - we KNEW we couldn't ever anticipate new apps or
patterns of use. At the time, computer-computer connections over LANs were
thought to be important. I had designed a protocol to "optimize" what we
thought might happen in those environments (as had PARC, with PUP). Rather
than "fork" the design to pursue our own egotistical desires to demonstrate
genius in a narrow subset of networking, both groups ended up supporting
the incorporation of a couple of the important insights into a
far-from-optimal Internetworking approach. And guess what? It paid off.
Lots of folks still believe that we are wasting computer cycles by not
writing code in hand-tuned assembly language. While true, that's wholly
irrelevant. The gain in portability, flexibility, and so forth is
thoroughly obvious when you step back.
Similarly, the success of the Internet (its flexibility and scalability
well beyond any of its '80s contemporaries) is thoroughly obvious -
goodness me, I'd just love to try to build a net that can support
commercial applications and services of this scale and diversity on the
X.25 networks or the ATM SVC networks. And if congestion turns out to be
better controlled by deploying twice as much fiber as needed, that seems to
me to be both theoretically and practically sound.
Hardware is now a commodity that is easily spent, something that
theoreticians have not come to understand in their framing of
"optimization" problems. The boundary conditions of any attempt to
optimize should incorporate the tradeoff between adding resources and
adding control complexity. [Students who want to productively rebel
against their teachers and textbooks would do well to focus on that one
fact. E.g., rather than optimizing the utilization of a link, determine
the optimum tradeoff between controlling the utilization and buying another
link, as a function of price.]
As far as competence is concerned, sometimes simplifying the problem and
focusing on real issues is the most competent thing one can do. Narrow
specialists often miss the big picture.
More information about the end2end-interest