On Oct 12, 2010, at 3:13 PM, Dave CROCKER wrote:
Bob Hinden and I chaired a working group that was answering your question
IPv6 was adopted and while there were a number of very different proposals.
The community chose to drop the work and ignore the issue for 10 or 15 years.
It happens that Deering's proposal came out of participation in our working
group, muttering something like "all this transition stuff is fine, but when
it's done, what we'll be left with will be ugly." So he designed his elegant
One bit of work that came out of the group was IPAE. More generally, it's
interesting to review documents of the competing proposals and note quite a
references to transition:
My point, here, is that the failures here were ones of goals, priorities and
management, not technology. Quite simply, we did not pay attention to larger
issues such as market incentives and adoption barriers.
Or people cared more about making the end result "right" (for some meaning of
"right") than about how to get there. (which is pretty close to the definition
of second system effect.)
There are a lot of things I didn't/don't like about HTTP, but I had to admit
that early versions (especially 0.9 and 1.0) were optimized for deployability.
Say what you will about its quality, efficiency, robustness, etc., but it's
certainly been successful from an evolutionary perspective. HTTP 1.1 tried to
fix a lot of the omissions in 1.0, and was partially successful, but it would
never have succeeded as an initial version.
I also remember that the current Internet email system evolved from a
hodgepodge of dissimilar systems, first by most of those systems adopting
(more-or-less) a common message format (at least for external traffic), then by
MX records providing a way to tie all of those dissimilar systems into a common
addressing framework, then (finally) by widespread adoption of IP and SMTP.
At each step there were incentives to local adoption. RFC 822 headers gave
systems a more flexible way to represent message metadata (especially with
things like Reply-To), MX records solved the problem of how to route traffic
between the Internet and mail domains not connected via IP, and moving to SMTP
provided faster and more-reliable service.
I've often wondered whether we could have used something like IPAE as a
stepping-stone to something like SIPP (which evolved into IPv6). Especially
since, in hindsight, it turned out to be much easier to get the host support
for IPv6, and even support for IPv6 in major applications, than to get the
Ietf mailing list