OSI and TCP Models


Fitch@DOCKMASTER.ARPA
Wed, 14 Sep 88 01:18 EDT


> Perhaps the best starting point for the discussion of a new
>model is with the successful implementations that work well in spite
>of the "official" model. What can be said about the model that these
>protocols actually represent? Of course, that's the way the tcp/ip
>"model" came about. It wasn't necessary to "codify" something except
>in response to another group that said "Here's our model."

> None of what I say is new. I was thinking that there might be
>common ground among the OSI layer model, the tcp/ip articulated model,
>and some new elements learned lately to fill in the obvious gaps.
>Something that can serve as a vocabulary and grammar, perhaps.

Kent England's comments reminded me of a few things. First, models tend to
come in two flavors: models of perfection and models of reality. Models
of perfection describe something that is a desirable goal to build to or
provide some common basis of understanding. Models of reality start with
measurable or observable things and are created as a way to describe what has
been observed and to predict some outcome.

It's sort of a chicken and egg thing. The ISO OSI model is more like a
model of perfection and if the TCP/IP model came about as Kent implies, then
it is more of the model of the second type. Neither is right or wrong; but
I think there is an underlying pr[oe]mise as to what one expects from a model.

While I'm at it, may be I'll stir the pot a bit more. Suppose the OSI
description of the Network layer says it provides protection from
unauthorized disclosure and routing. Suppose our network protocol NP
provides these services by invoking various key distribution protocols,
address resolution protocols, ... as well as functions actually performed by
the NP protocol itself. As far as layer N+1 (transport in this example) is
concerned, the Network Layer comes through with the desired services. But
can we say that NP conforms to the model? NP is a single protocol. NP
coupled with address resolution protocols, key distribution, ... perhaps
conforms. The point is that it is probably not a good or fair
question/statement to say a single protocol conforms to an OSI style model.

As an aside, a few years ago when I was at MITRE, we created a model that
distinguished between user related protocols, management protocols, and
interface protocols. This made a three pronged swiss army knife model.
I suppose this means that anyone can create a model and nail it to a church
door, but it's the size of the following that counts. This probably also
explains why we get religious about models.

--Jack

Fitch@Dockmaster



This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:43:30 GMT