[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ih] Fwd: [IP] OSI: The Internet That Wasn't

> IEEE Spectrum just published online, titled "OSI: The Internet That
> Wasn't."  OSI, of course, is the acronym for Open Systems Interconnection.

The article's characterization of the IETF's activities in 1992 is, 
itself, a gross mischaracterization of what took place.  I was on the 
IESG at the time and my own reading was that CLNP had a reasonable shot 
at being selected from amongst an array of contenders.  However the 
IAB's premature selection of it -- rather than letting the community 
continue through an evaluation process -- finally brought to a head the 
continuing tension between the IAB's style of exercising authority and 
the community's festering frustration with it.

More generally...

The author interestingly entirely missed the real lessons for why OSI 
failed.  As noted in some of the other postings here, TCP/IP had 
deployed useful code and OSI did not, except in very small scale and 
very strict operating conditions.  Also it was, indeed, massively more 
complicated than the Internet stack, but it also was incomplete.

In the late 1980s, I managed various efforts at developing commercial 
TCP/IP stacks.  This also included an OSI stack.  When we started to ask 
customers about their needs for our product support in transitioning 
from TCP to OSI, they said that what they actually needed was support 
from OSI to TCP.

Perhaps as much as 25% of my customer base was in Europe, the hotbed of 
OSI.  OSI created the market demand; TCP/IP satisfied it. (This is a 
commercial version of Stef's quoted comment.)  One of my customers 
was... ISO, the home of OSI.  I asked their IT manager whether he got 
criticized for supporting TCP/IP and in typical operations manager 
humorless pragmatics, he said "they gave me an operational requirement 
and this was the only way to satisfy it."

In the early 1990s, I wrote some articles on the Internet standards 
process and tried to describe the apparent difference in the IETF's 
engineering style vs the world of OSI.[1]  What I finally concluded was 
that both communities had serious, bright engineers who were trying to 
do good things.  And both communities had engineers who would each bring 
long and different lists of requirements.  The distinguishing 
characteristic between Internet and OSI engineering was/is how the sets 
of lists were processed.  In the OSI world they would try to satisfy the 
union of the lists.  This demands a large complicated system that takes 
a long time to produce.  The Internet looked for the intersection, 
thereby permitting earlier delivery of an essential subset.

And that, I believe, is the actual core lesson from this history:  For a 
complex problem space, find a useful subset that can be delivered 
quickly.  Deliver it and start gaining field experience.  Based on that 
experience, then start extending the capabilities.


[1]   http://bbiw.net/musings.html#standards

Dave Crocker
Brandenburg InternetWorking