[ih] GOSIP & compliance

[Alex McKenzie]

 I'm moved to add my $0.02.  I was deeply involved in both ARPAnet standards (I was the editor of the ARPAnet NCP document), and I was chair of the ISO Session Layer subcommittee.  From those perspectives I make the following observations.
1. ARPA/DARPA provided the money for the first users of ARPAnet and later the TCP Internet.  ARPA demanded that they work, and soon.  The protocol design committees felt that pressure.  The ISO committees were mostly made up of industry representatives who wanted their own private protocol stack incorporated into the standard and were willing to delay approval until that happened.
2. ARPAnet and the TCP Internet were intended to ensure that every connected system could connect to and interwork with every other connected system.  If there were to be options in any protocol, they were options to be added to the basic protocol implemented by everyone, and options could be refused.  OSI was intended to allow manufacturers and software vendors to certify that they were compliant.  For example, at the Transport layer there were several options; no option was required and the options were not interoperable.  So the CCITT Transport, the GOSIP Transport, the IBM Transport, etc could all be compliant and not interoperate with each other.  As another example, the ARPAnet/TCP Telnet protocol demanded that every system be able to send text in the ASCII encoding.  In the ISO Session layer meetings, many participants demanded that there be NO minimal standard text encoding - rather there should be negotiation about what encoding would be used, with the possibility of finding that two systems did not support any common encoding and yet both were compliant.
3. ARPA/DARPA enthusiastically supported the development of the "host" software implementing the ARPAnet and TCP Internet protocols for a wide variety of the most common computers, starting with the mainframes of the early 1970's, thru the workstations of the 1970-80's, and to the personal computers of the 1980s.  ARPA/DARPA encouraged this software to be made widely available cheaply or for free.  The OSI software development was not supported financially by governments, it was expected to be developed privately by computer manufacturers (in addition to their own proprietary protocol software), or by software vendors who could recover their development costs through expensive sales or licensing.

4.  As pointed out by John Day, a lot of the OSI work was a battle between the PTTs and the computer manufacturers about who was going to "own" the added value that networking would provide.  The PTTs wanted a protocol architecture that kept the added value in the network.  The manufacturers wanted a commodity network with the value outside.  For TCP, ARPA had settled the argument in favor of the computers.
In view of these factors, it is not surprising that when people wanted the values networking could bring (and especially after the development of point-and-click web browsers) they opted for TCP.
Cheers,Alex

 
 On 3/19/2022 7:59 AM, Clem Cole via Internet-history wrote:
> *IP vs. OSI    -- "**Simple Economics always beats Sophisticated Design"*


This is certainly an appealing saying, and it might even be true. 
Sometimes.

But it does not describe the core reason OSI failed and TCP/IP 
succeeded.  By the time this saying was relevant, TCP/IP had already won 
the war.  Rather, this saying merely describes coming to the recognition 
of which won.

That real core was more like: "simple, operational technology always 
beats elaborate, incomplete, dysfunctional technology".

OSI was /not/ sophisticated design.  It was cumbersome /over/-design.

The reference to TCP vs. TP4 is an example of missing the point, since 
there was a mess of other TPs, for use depending on what the underlying 
networking technology was.

For the Lynch & Rose 1993 book, Internet System Handbook, I did a 
chapter about Internet technical processes, which prompted my 
considering differences between Internet and OSI processes.  (I had some 
limited experience in the OSI realm.)

Simply put, I believe the two communities did not differ in 
intelligence, knowledge or intent, but in pragmatics and a core bit of 
politics.  The OSI work required unanimity, which meant pleasing 
everyone, which meant including pretty much everything from everyone's 
various laundry lists.  This meant design took an extraordinarily long 
time, while tending to produce highly bloated specs.

In contrast, the TCP/IP community typically wanted something work by 
yesterday, which mean using only the intersection of everyone's lists. 
That produced smaller designs, with an implicit basis for knowing what 
was included would be useful.

A revised version of that chapter was published as Making Standards the 
IETF Way
<https://bbiw.net/ietf/ietf-stds.html>
1993, Association for Computing Machinery [Reprinted from StandardsView, 
Vol. 1, No. 1.

There were, of course, a number of other differences that probably had a 
large effect, including meetings (open vs. closed), primary venue 
(online vs. f2f), and document access (free vs. charged).


d/
-- 
Dave Crocker
Brandenburg InternetWorking
bbiw.net
-- 
Internet-history mailing list
Internet-history at elists.isoc.org
https://elists.isoc.org/mailman/listinfo/internet-history