[ih] Why TCP?

[Dave Crocker]

> I think Noel is largely correct although it was the NSF decision to use
> TCP/IP for both CSNET and then NSFNET that drew significant academic
> uptake. The ARPA-sponsored academic sites were mostly CS research into
> AI and related areas - numbering around a dozen in the early days, but
> growing to about 40-50 as the TCP research continued.


Large-scale adoption of innovation is a sufficiently social process to 
make precise accounting impossible.  Still, it might be worth listing 
the serious of choices and factors people consider as likely 
contributing factors at different levels.

A quick start:

In terms of design, the focus on relative simplicity and consistency 
(one TCP rather than 5 OSI TPs, for example) and the layering, such as 
separating TCP and IP, dramatically reduced technical learning curves. 
(I suspect so did the tendency to use textual encoding rather than binary.)

The very long history of practical field experience, with 10 years of 
production use before it started to hit the larger mass-market.  Besides 
making for well-baked code, it ensured end-to-end operational usability. 
  The DNS was not a small part of this.  The competing OSI effort 
largely failed to ensure complete solutions (or at least failed until it 
was far too late, but I tend to think they never got there or when they 
did, had largely unworkable solutions, such as email addressing.)

Starting in the more constrained environment of R&D might hurt some 
kinds of technical deployment but in this case it allowed an 
infrastructure to grow independent of most national and business 
push-back, until much later in its life.

While the DARPA effort created the core technologies and its initial 
operational use, NSFNet's business and operational model pushed for a 
number of developments that became critical for the large-scale growth. 
An inter-organization routing protocol that did not rely on a single 
backbone.  Regional providers that seeded the commercial market later.

There were other critical decision points, such as the CSNET/NSFNet one 
Vint cites.  Craig Partridge also note the switchover to common use of 
domain naming among the various, independent messaging networks.

Feel free to add/modify...

d/

-- 

   Dave Crocker
   Brandenburg InternetWorking
   bbiw.net