NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/protocol/tcpip/6134

Text File | 1993-01-28 | 5.3 KB | 104 lines

Newsgroups: comp.protocols.tcp-ip
Path: sparky!uunet!ukma!darwin.sura.net!sgiblab!newsun!donp
From: [e-mail] (don provan)
Subject: Re: Info wanted on TCP/IP vs OSI 7 layer
Message-ID: <[e-mail]>
Sender: [e-mail] (The Netnews Manager)
Nntp-Posting-Host: na.sjf.novell.com
Organization: Novell, Inc., San Jose, California
References: <[e-mail]> <[e-mail]> <[e-mail]>
Date: Thu, 28 Jan 1993 20:15:34 GMT
Lines: 91

In article <[e-mail]> [e-mail] (Ketil Albertsen,TIH) writes:
>Sorry, I don't understand what you mean by "layers turned sideways".
>Could you explain? (Preferably with examples)

Actually, you've been clearly illuminating the most obvious example:
x.25 support.  X.25 was wedged into the OSI model by inventing a null
transport layer and a null network layer so that the datalink layer,
x.25, can manifest itself directly at the transport service layer.
One way to envision this is as turning the datalink, network, and
transport layers on their sides so that datalink can reach up to the
bottom of the session layer.

Another example involves the upper three layers.  In practice, these
don't really work as "layers", but more as individual units in a
cooperative whole.  In any practical implementation, the actual
application (as opposed to the application layer) has to interact with
presentation and session directly, not through any higher layer
service exchange.  This is reflected in the service definitions, which
include many "pass through" type functions in the presentation and
application layers, although not really enough to do a reasonably
efficient implementation.  Again, the way i've started to envision
this is by tilting the three layers on their sides such that the
application can interact with any of them directly.

And, of course, with the introduction of ACSE, the application layer
itself has become more of a utility library than a layer.  This is a
different kind of tilt: while the model envisions various application
protocols in the application layer, each running vertically from
presentation up to the top of the stack, ACSE is now the one thing
that goes from top to bottom, dealing with each of the application
protocol components as appropriate for the actual application.  In
effect, the protocol components hang out horizontally from the ACSE,
each connected only to the ACSE rather than spanning the space between
the application and the presenation layer as dictated by the model.

I hope this clears up what i meant.  Bear in mind that these aren't
criticisms.  So far as i can see, these are the *correct* solutions to
the problems encountered, not deficiencies in the protocol designs.

>Sure. The protocols came first, before a good taxonomy had been developed.
>So the OSI guys had to face a decision: Should the taxonomy, the RM,
>describe the world in the current messy state, or should we make an
>attempt to be systematic, create a goal to work towards? They chose the
>latter, and I think that was a good idea.

I do, too.  My point, though, is that it is a *taxonomy* (an excellent
term for it: thank you for mentioning it).  As a taxonomy, the model
is good only for *describing* protocols, not for analysing or
evaluating them.  And, as i say, this is revealed over and over
whenever someone tries to "compare" the two protocol suites using the
OSI model as the measure of quality by which they are rated.

>(A sidetrack: If your statement was intended to say that TCP/IP is well
>layered, I think most datacomm people would protest!

Well, i intended to say no such thing, but in my experience -- and
forgive me for saying this -- most datacomm people wouldn't know
layering if it jumped up and bit them on the nose.

>I think you could day that the more recent OSI protocols are generally
>closer to the model, but there are certainly some problems even in the
>newer ones.

As i say, i think it would be a mistake to measure the quality of a
protocol suite based on how close it comes to matching a taxonomy.
That would be something like judging the quality of a cow by measuring
its "cowness" rather than by measuring how much milk it produces.

>I do believe that in ten years, we will smile at the current low layers
>in OSI (and that is NOT because IP will take over... :->).

I'm not sure where you're looking, but from what i can see, in ten
years we won't even remember OSI.  With all the excitement about the
growing OSI market, everyone seems to overlook the fact that the
TCP/IP market is growing an order of magnitude faster.  IP doesn't
have to "take over", since it's the installed based in virtually every
market that isn't dominated by proprietary networking.  From
everything i've heard, the only people still interested in OSI are the
ones that have to deal with painful, government induced network
monopolies such as the European phone companies.  (I've lost count of
the number of times i've heard about people that wanted OSI only to
connect geographically separated *TCP/IP* LANs.)

>I do not feel "bothered" by a discussion of Internet vs. OSI protocols,
>even if what started the discussion was a question from a student.

I'm not bothered by the discussion, either, although this discussion
actually has very little bearing on the original question.

                        don provan
                        [e-mail]