When we reported the awarding of the Turing Award to Vint Cerf and Robert Kahn for the TCP/IP protocol, we had a lot of Unix greybeards commenting and reminiscing. Unix historian Peter H. Salus left the following comment:
"It's important to recognize, while we laud Vint and
Bob, that *all* the work on the ARPANET was open and
unencumbered by patents, etc.
"Steve Crocker, Ray Thomlinson (mail), Dave Walden,
Len Kleinrock, Jon Postel, and nearly 100 others
get credit for these wonders, which I'd not even
thought about before the mid-1970s.
"There was a decades-long battle between OSI and TCP/IP,
with all the manufacturers and the PTTs lined up
against TCP/IP. TCP/IP won because it worked, because
it had been implemented, and because the networkers
actually used it.
"Bouquets to Bob and Vint are well-deserved. For over
three decades of work each."
We also then had a great suggestion from iceworm, seconded by several, including me, that Salus, who is a Groklaw member I am happy to say, might like to write a more complete article for Groklaw, to fill in some of the blanks in the story from the New York Times.
"Perhaps it would be well to unravel the 'OSI' stack and why it did not become the standard for the Internet. I dimly remember discussing it in one class meeting (well, part of the class) in the semester I took a course in networking about five years ago. I remember (dimly) some comments along this same line of why TCP/IP won the day, but I don't remember the details.
"We did use the "OSI" concept to make short presentations. We never did examine the differences between the "OSI" stack and the TCP/IP concept. Again, perhaps there is someone who could describe these differences in layman's language which might be of interest to the Groklaw community.
"Finally, I read a little on the concept behind IBM's OS/2. There seemed to be some bit of the "OSI" concept here also in that there were "application" and "display" layers, if I remember correctly. So a further question comes to mind, namely, "Did OS/2 become a niche player because it was closed and could not compete with the monopoly OS, or was its underlying design deficient as compared to the Unix kernel design?"
Peter graciously agreed to write an article giving us the history, and here it is. I love the barfbag detail. For the detailed comparison of stacks
that iceworm wanted, those are found in Salus' book, "Casting the Net: From ARPANET to INTERNET and Beyond
". Please feel free to add your own memories here, so it will be a permanent record for our enjoyment and for historians in the future.
Some are reportedly thinking of writing a letter of formal complaint to the Association for Computing Machinery (ACM), which gives out the Turing Award, because of perceived issues with MCI, Cerf's employer, and spam, protesting the award to Mr. Cerf. I hope they don't embarrass themselves with such an ungrateful, misguided, and unkind gesture. Spam is awful, but it is not, I must say, the only awful thing people can do to one another. Internet thuggery is plenty awful too. Mr. Cerf should be allowed to enjoy this wonderful honor, which is well-deserved, in peace. We ought to all be allowed to disagree with each other on issues of concern to us all, without being attacked and painted in evil brush strokes. And there truly is no connection between this honor and spam. It's just not nice to ruin this moment, and I hope they will reflect and desist.
OSI and TCP: A History
by Peter H. Salus
[The following is an abbreviated conflation of pp. 117-126 of my
Casting the Net (1995) and my review of Padlipsky, The Elements
of Networking Style (1985; 2000) to appear in //www.unixreview.com.]
In 1977, the British Standards Institute proposed to ISO, the
International Standards Organization, that a standard architecture
was needed to define the communications infrastructure. (As I've
said and written repeatedly, this, as with IFIP, CCITT, and other
efforts, shows how the road to hell is paved with good intentions.
Because X.25 was unsatisfactory, the IFIP Working Group was set up
in the vain hope that the technical community could block the political
arena that is ISO. It didn't.)
ISO set up a subcommittee of a technical committee to study this
(ISO/TC 97/SC 16). The next year (1978), ISO published its
"Provisional Model of Open Systems Architecture" (ISO/TC 97/SC 16 N 34).
This was labelled a "Reference Model," and referred to as OSIRM (Padlipsky,
my favorite curmudgeon, labelled it ISORM -- "eye-sorm").
In general, ISORM was based on work done by Mike Canepa's group at
Honeywell Information Systems, which came up with a seven-layered
architecture, which itself owed a lot to IBM's proprietary Systems
Network Architecture (SNA). SNA had been announced in 1974 and its
seven layers do not correspond exactly to ISORM's.
TC 97/SC 16 turned over proposal development to ANSI, the American
National Standards Institute, to which Canepa and his technical
lead, Charlie Bachman, presented their layered model. This, in turn,
was the only proposal presented to the ISO subcommittee in March 1978.
It was accepted and published immediately.
A "refined" version of the ANSI submission to ISO appeared in June
1979. This published version is (gosh!) nearly identical to that of
Honeywell in 1977.
Dave Walden and Alex McKenzie had pointed out that both virtual
circuit and datagram services were valuable. "An international
standard would do well to support both," they wrote. The 1977-1979
models were such that the extant host-host protocols didn't fit.
To me, there was a clear opposition: ISO was trying to construct
a nice set of geometric figures that would be a "tidy model"; the
ARPANET folks were interested in getting things to actually work --
they wanted to push bits around the system.
The irascible Padlipsky has described the OSI model as a pair of
high-rises with parking garages. The basic model had a pair of
seven-storey buildings; reality complicated things.
John Quarterman pointed out a decade ago that "OSI specified before
implementation. So specification took forever and implementation
never happened. ... OSI and IP started about the same time (1977).
OSI wandered off into the weeds and got lost. IP won the race."
There are two more things to point out.
The EC (now the EU), the European and the Japanese
PTTs all supported ISORM. They pushed hard for it in a variety
of ways. But, by the end of the 1970s, TCP/IP was a suite with a
real installed base. Those implementations had evolved between
1974 and 1978 and had become accepted by the technical community.
By supporting a theoretical specification with no implementation,
the Europeans and the Japanese became identified with the "old"
telephony and telegraphy.
As all the ARPANET work had been supported by the US DoD, the
EC and Japan appeared to mired in NIH -- not invented here. And
the European and Asian governments feared that this would give
US manufacturers a boost.
Finally, as the ARPANET (soon to become the Internet) had been
governmentally funded, it was (as we would now say) freely
But real money was involved. Future profits were at stake.
The result was that many companies wasted vast sums of cash for a decade
trying to produce OSI products while the networking community
moved ahead with TCP/IP.
As early as 1986, Jack Haverty and Gary Tauss were sure that the
market had "the choice between a well understood, widely implemented
protocol set, namely the Internet Suite, and an evolving, incomplete,
not widely available set, representing the OSI model" (Government
Data Systems 15.3, April/May 1986).
The fact that local area networks and the manufacturers of desktop
equipment supported TCP/IP meant that there was an ever-waxing
infrastructure built on TCP/IP in Europe. The changing situation
in Eastern Europe and in the (then) USSR, meant ever-increasing
opportunities for TCP/IP, as all the protocols were in the public
domain and thus no fees were payable.
By a dozen years ago, while there were some OSI networks, there was
no growth. Store-and-Forward networks like Bitnet were shrinking,
and the growth of Fidonet and UUCP had slowed dramatically.
The US DoD played no small role in proliferating confusion. I don't
want to go into either AUTODIN (Defense Information Network) nor
AUTODIN II here. But, in its customary fashion, the Department let
contracts to all the usual suspects: Western Union, Ford Aerospace,
CSC, and Mitre all received contracts. "Western Union knew how to
talk to the military," Dave Walden told me over a decade ago. "We
didn't. For example, they wanted security. We [BBN] said: 'This
is easy. This is hard.' Western Union said: 'Of course we can do
that.' So they got the contract."
Padlipsky, I was told on good authority, once marched into a
meeting with the vendors, pulled an airline "barfbag" from his
attache case, put it on the table and said "OK, you guys, I'm
ready for you."
It turned out that Western Union couldn't do the job. The DoD
finally convened an "execution squad" and killed the project. It
had taken over 15 years and untold millions of dollars, but finally
Vint Cerf, Bob Kahn and others persuaded the military brass that
the ARPANET/Internet protocols reliable, available and survivable.
I've been told that it was Frank Carlucci himself, when he was
Secretary of Defense (1987-1989), who actually signed off on
terminating the Western Union agreement.
Over the decades, the National Bureau of Standards (now the National
Institute of Standards and Technology) issued Federal Information
Processing Standards, which set down the requirements for vendors
interested in purveying to the government (an enormous market).
In the late 1970s, a FIPS was issued requiring OSI capability. So
manufacturers dutifully supplied buttonholes and velcro, should
anyone ever actually implement ISORM. Only in 1994 was the FIPS
I think this is relevant and important because, though it took
over 15 years, technology won. The "good guys," led by Cerf, Kahn,
Walden, and others defeated the governments, PTTs, and giant
corporations by using well-engineered, open protocols -- stuff
that worked and was robust.
Copyright © 2005 Peter H. Salus