[e2e] the evolution of deployability
David P. Reed
dpreed at reed.com
Wed Dec 18 08:12:00 PST 2002
At 12:00 AM 12/19/2002 +1100, grenville armitage wrote:
>Only the naive think standards bodies exist for the betterment of
The great thing about standards is that you have so many to choose
from. Standards bodies don't like that, but have to live with it, until
they get to play the lock-in game.
IPv4 was a great standard. It didn't specify routing and it didn't specify
higher level protocols. That was the beauty of the end-to-end argument.
IPv6 looked like a good upgrade to IPv4. It dropped fragmentation, and
still didn't specify routing and didn't specify higher level protocols (it
only changed address sizes and created opportunities for better
encapsulation). Pretty simple.
Then somebody decided that IPv6 needed new routing protocols before it
could be rolled out. This should not have been an issue - routing
protocols were being rolled out with no change in IPv4 whatsoever, so it's
clear that the hourglass model made this an independent decision.
This idea (that all layers are interdependent) is the source of the idea
that "IPv6" was not ready.
Rather than recapitulating the success of the Internet based on the
end-to-end principle, IPv6 had been invaded by ISO/OSI bellhead thinking.
Books were being written about IPng and IPv6 that focused everywhere but on
the neck of the hourglass.
Creeping featurism took over.
The last time I saw this was at IBM, when IBM devoted 30% of the company
resources to a system called FS, which was designed to be the generation
after the 360/370. It was an all encompassing comprehensive design,
based on all the latest thinking in all areas of CS.
The end result was a system that could not be delivered, met no customers'
needs (but ALL of their requirements!) and needed to be shut down at great
cost to the organization.
Where was the simple IPv6 that was designed to be a small upgrade by its
The name had been "colonized" by people with other agendas.
IPv6 should be called the "anything but IP" project. Major offenders were
those who tried to embed new router technologies in it, although I'd add
high marks for those who tried to force their dreams of QoS into it as
well. But the worst were those who saw NAT and other middleboxes coming to
put twisty passages into the network, and did nothing to compete for those
real customers with real money and real needs. (ARIN could have
encouraged IPv6 as a way to conserve address space if IPv6 overlay
networking were allowed to flourish).
More information about the end2end-interest