[ih] Ken Olsen's impact on the Internet

[Dave CROCKER]

On 2/9/2011 1:28 PM, Richard Bennett wrote:
> The general understanding among computer companies in the mid-80s was that
> TCP/IP was a fine proof-of-concept, but the real network was going to be OSI.

I believe that was also the expectation within the Internet technical community, 
even to the point of there being efforts to hand TCP/IP over to the OSI folks. 
(Who rather rudely rebuffed the offer.)


> This wasn't any sort of conspiracy as much as it was a recognition that large
> scale networks needed a different kind of system for addressing and routing than
> the one that IPv4 provided, and that TCP would have problems on fatter pipes.

This implies both a more careful understanding of the problem and solution 
spaces, as well as a sufficiently deep understanding of TCP/IP's limitations, 
than actually took place.  In other words, what you are citing was common 
rhetoric but had no substance, in my observation.

As for addressing and routing, the OSI world eventually produced something 
useful for /interior/ routing, but never for inter-organization routing.  So 
whatever the claimed concerns, after 15 years of effort, the OSI world produced 
nothing viable for Internet scale addressing or routing.  As with most OSI work, 
the deliverable of field utility was always two years from now.

For email addressing, the OSI model chosen was actually unworkable at at scale. 
  For the same mailbox in your organization, you needed a different public 
address for each provider (common carrier) that you were connected to.  This 
made for some amusing, if quite silly, business cards.


> The OSI development process was ultimately unsuccessful for a number of reasons
> (too many cooks, counter-lobbying by IBM, the clambering of the European PTTs
> for connection-oriented systems, the lack of any real champions, etc.) so the
> networking industry was left to do the best they could with TCP and IP.

There might have been efforts within the OSI world to defeat OSI, but everything 
I saw from the outside says quite the opposite.  Industry and government 
commitment to OSI was massive, to the level of religion.

Rather, what I saw were two core, strategic errors.  The first was horrendously 
complex, interdependent technology components and the second was a failure to 
understand the need to obtain real-world operational field experience quickly 
and base revisions on it.  (Deploy something useful as quickly as possible and 
grow the service technology from the experience.)

The error on the technical side was pretty classic "big system syndrome" along 
with a failure to adequately understand end-to-end interoperability 
requirements.  Observe, for example, the number of different and 
non-interoperable connection-based transport protocols seeking to provide 
essentially the same type of service to the client layer (TP0-TP4).  The premise 
of trying to optimize for different underlying network environments is quite 
natural but proves fatal in this service space.

(As I understand it, the TCP effort had a close call with this same issue, when 
LANs started to be popular.  I heard there was strong pressure to have a version 
of TCP tailored LANs but that Vint vetoed it.)

 From a design standpoint, there is a classic tradeoff between universality 
versus (local) optimization.  For an integrative, large-scale service, the 
former has proven far, far more important than the latter.

The big system syndrome meant that it was not possible to get essential 
operational experience early and learn from it.  (For reference, the biggest 
contribution to OSI field experience for applications came from the Internet, 
with Marshal Rose's OSI Application-over-TCP package, ISODE [RSC 1006].  So much 
for claims the Internet was hostile to OSI...)

  d/

-- 

   Dave Crocker
   Brandenburg InternetWorking
   bbiw.net