[ih] NCP to TCP/IP Transition

[Jack Haverty]

Matthias,

I still somewhere have the button which says "I survived the TCP
transition"!  

In order to understand the transition, I think you might want to look
into the context, especially other concurrent activities.  In
particular, the US Defense Data Network (DDN) was coming into being at
the same time.  The Arpanet mission was research, but over the years a
number of hosts had come onto the Arpanet from the operational
community.  In fact, DCA had by then taken over the role of operating
the Arpanet, reflecting the fact that it had evolved to be more
operational in nature instead of a research tool.

In the original long-term plan, a network called Autodin-II was supposed
to provide the future operational communications needs.  Autodin-II
would have been built by the traditional contractors, using approaches
more associated with PTTs and "phone companies" than with the new
technologies of the Arpanet and TCP.  However, the government did a
study, and decided to cancel the Autodin-II efforts and replace them
with new programs using the Arpanet technology.  I was working at BBN
during those years, and I remember we were all stunned to learn that we
had actually won the competition to provide the new network(s) using the
proven Arpanet technology.  

Vint's efforts for several years before the 1983 date undoubtedly helped
prove that the Arpa technology was ready to deploy.   I remember that we
(in the research community) were protesting that the TCP technology
wasn't done yet.  At every meeting, there were always many items on the
"things to be figured out" list.  But of course a researcher is never
done - there's always another aspect to investigate, or refinement to
make.  Don't take my toys away!

I googled quickly and found a short summary of some of those events:

http://www.chips.navy.mil/archives/87_Jan/FILE22.HTM

There are no doubt many other such discussions.

Since by then it was also clear that local networks of various
technologies had to be supported, TCP was required.  NCP was not
designed to interconnect networks.  Among other things, this also meant
that the relevant protocols (TCP, IP, etc.) had to become "official", so
there was considerable work done to establish them as official "DoD
Standards".  Unlike the research environment, the operational world
demanded more structure, more control, more paperwork, and more
procedures.  With TCP/IP as official DoD Standards, they could then be
used in procurements for operational programs - things like Army payroll
systems, etc. that had nothing to do with research.

The Arpanet NCP/TCP transition was one piece of this overall migration
of the Arpa technology research product into the operational government
"customer" world.

As I remember, DCA pushed very strongly on this transition.  After all,
the Arpanet was in effect the beta test site for the subsequent
introduction of TCP to the operational community which DCA served.  It
was important to make sure everything worked in the Arpanet to avoid
problems in the operational world which is much less tolerant than a
research community.  Imagine trying to tell your Army that they didn't
get paid this week because of a TCP problem.

I believe that the DCA pressure came from the adoption of TCP as the
future (then) technology for the entire operational community.  If that
decision to switch from Autodin II to DDN hadn't happened, there
wouldn't have been any pressure, and the transition to TCP probably
wouldn't have happened. 

The money that was "poured into TCP implementations" was insignificant
compared to the money to be poured into various applications that were
required to change to use TCP (since it was DoD Standard).  That's where
the pressure came from.

Although there was a particular date/time when the "transition"
occurred, I think it's misleading to think of this as a "flag day".
After all, the Arpanet was always able to carry NCP and TCP
simultaneously.  Hosts had a long time to prepare, and could even choose
to operate both NCP and TCP at the same time, and the "application level
gateways" made it possible for the separate NCP and TCP worlds to
continue to communicate as each pursued the transition at their own
pace, as long as they completed it by 1/1/1983.  The email
interconnection was by far the most important and as I remember it
worked fine.

January 1, 1983 was simply the date on which NCP would no longer work.
It was a simple patch to the IMP code to make NCP fail, and it was
easily reversible (but that was of course not advertised).   So that
date put pressure on all the hosts to do what they had to do in order to
continue to function after the switch to TCP.

I wish I could remember more about the cutover.  For example, how much
NCP traffic was still running immediately before NCP was disabled?   As
I remember, the transition was pretty much a non-event.  No disasters,
no frenzy to fix problems. 

One way to think about all the events around that time is that it was
all part of a massive multi-year "technology transfer" of ARPA-developed
research technology into the "real world" of the government
operations.  

Through the late 80s, a similar technology transfer started, introducing
TCP into the commercial world, with early adopter corporate internets
being deployed using TCP.  This was the death knell for competitors -
DECNET, XNS, SNA, X.25, etc.  

Then I would argue that there was another later "technology transfer" of
the TCP technology into what people now call "The Internet", when the
Web came into being and exploded and made the contemporary "public
networks" (AOL, Compuserve, etc) wither.

Hope this perspective helps,
/Jack Haverty
Point Arena, CA, US


On Mon, 2009-04-27 at 12:50 -0400, Matthias Bärwolff wrote:
> Hi,
> 
> is my understanding correct in that it actually took more than a year to
> do the transition from NCP to TCP/IP, starting roughly in early 1982 and
> being done with by mid-1983? This is what RFC 801 (the plan), and rfc842
> through rfc848 plus rfc876 (the progress reports) seem to indicate.
> 
> I am asking because the transition is nowadays always being referred to
> as a flag day transition (which in may understanding is defined as the
> very absence of any transition period, e.g. something like changing from
> driving on the left to driving on the right side of the road), but
> apparently this was neither the case, nor was it intended to be. There
> was simply a deadline, which, of course, was not met.
> 
> Also, I am wondering, have the application layer gateways (relays) that
> RFC 801 refers to been deployed, and if yes, to which extend, and how
> successful?
> 
> And a final question, while I'm at it: How decisive was the pressure
> from DCA in this? My impression has been that without the top-down
> pressure the whole thing may well never have happened, despite all the
> money being poured into TCP/IP implementations.
> 
> Pointers to relevant literature sources are appreciated, too. Thanks
> again for your help.
> 
> Matthias
>