Evolutionary Approaches to Revolutionary Change in the Internet Architecture
============================================================================

In the past several years, the networking research community has grown
increasingly frustrated with the difficulty of effecting substantive
change in the Internet architecture.  This frustration is responsible,
in part, for a shift in focus to research in "green-field" environments 
such as overlays, peer-to-peer, sensor networks, and wireless networks.  
Although these topics are compelling and important in their own right,
I would argue that the underlying Internet infrastructure, IP protocols, 
and operational practices are plagued with serious problems that warrant 
significant research attention from our community.

However, having revolutionary impact on the Internet architecture is
extremely challenging because of the commercialization of the network
and the large installed base of legacy equipment running the existing
protocols.  As a result, many studies of the Internet's "underlay"
focus on collecting and analyzing measurement data or on proposing and
evaluating incremental changes that address specific problems.  These
kinds of studies are certainly interesting and important, both to
strengthen our understanding of the current architecture and to
improve the performance and reliability of today's networks.  Yet, we
need to find ways to effect more fundamental change in the years ahead.

I think that breaking the stalemate requires us to think creatively
about evolutionary approaches to revolutionary change.  The key is the
sometimes tedious notion of "incremental deployability" (an important
concept that is often abused as a weak excuse to reject papers that
take green-field approaches, or to criticize papers on point fixes for
their preoccupation with today's artifacts), which should give way to
a deeper treatment of how to exploit the evolvability of the existing
protocols and systems, and how to make them more evolvable in the future.
To effect change, I think we should consider the following three issues:

  - Target end state: We need to think deeply about what we want the
      Internet architecture to look like in the future.  This may (and
      arguably should) include empirical studies of the operation of
      today's networks and protocols, discussions of the requirements
      for our future infrastructure, and green-field investigations of
      new architectures.  A focus on identifying and addressing the
      foundational issues should dominate the agenda rather than point
      fixes to specific sub-problems, though some useful point fixes
      may certainly arise along the way.

  - Backwards compatibility: To effect change, we need to find ways to
      move toward the target end state without requiring a "flag day"
      where the legacy equipment and protocols are replaced.  For
      example, approaches that maintain the message formats of our
      existing protocols can allow substantive change while still
      accommodating legacy equipment.  Ethernet is a great example of
      this principle -- significant change has occurred in the past
      decade or so while (and arguably *because of*) remaining
      compatible with the legacy specification of the message format.

  - Incentive compatibility: Each step along the way to the target end
      state needs to offer substantive benefits to the early adopters,
      and additional incentives for the remainder of the Internet to
      join the party.  The need for incentive compatibility is so strong
      that it might dictate the choice of steps and, in some cases,
      even the target end state itself.  A sound treatment of incentive
      compatibility requires accurate models of the cost of deployment
      and the benefits at each stage of adoption.  The construction of the
      steps along the way is arguably a research problem in its own right.

To make progress in this kind of thinking, I believe the networking
research community would benefit from engaging at least the following
two specific communities:

  - network practitioners (operators, equipment vendors, campus
      administrators) to understand the existing protocols, the
      limitations of today's architecture, and to collect valuable
      measurement data; this includes networking researchers creating
      their own opportunities to collect data and gain operational
      experience by building and running large research testbeds

  - researchers with expertise in economic modeling, game theory, 
      and mechanism design to start developing ways to reason about 
      incentive compatibility and the competitive forces that have
      such a significant impact on the Internet's operation and evolution

There are undoubtedly other communities as well (e.g., complex systems
and network optimization) that would help us in putting the design and
evolution of the Internet architecture on a strong foundation and in
charting a course that can ultimately lead to fundamental change.