[ih] A comment on the seven layer model

[Bill Ricker]

On Tue, Apr 22, 2025 at 4:58 PM Steve Crocker via Internet-history <
internet-history at elists.isoc.org> wrote:

> I liked your comment, "I gave up long ago on trying to stuff this into a
> 7-layer diagram and explain it."
> ...
>
>From the beginning we thought in terms of thin
> layers that provided useful services, with the proviso that others would
> build on, ignore and build around, or insert other layers in between as
> needed


I don't have MAP's original canonical phrasing immediately to hand, but as
he happily republished my copy-pasta of John S Quarterman's paraphrase, he
had accepted it as close enough:

> I like Mike's statement, which I'll paraphrase
> "If you know what you're doing, 3 layers is enough; if you don't, 17
layers won't help you."

MAP also pointed out that the OSI ISORM's holy seven layers were _not_ all
end-to-end, though usually drawn that way: the bottom 3 layers were to be
traversed in both DTEs' attached DCEs, so in reality the 7-story apartment
buildings should always be drawn with shorter parking garages between them,
with full tour of each parking garage required.

(If one searches, one can find OSI diagrams with 7 layer DTEs left and
right with a pair of 3 layer DCEs betwixt, but it's rare, outside of
materials regarding e.g. DataLink layer or Frames. This is not to imply
that non-OSI systems didn't often have DCE with similar constraints, but it
wasn't hard-layered or required.)

(And MAP also observed that hard layering of Session "below" Presentation
prevented sensible engineering in ways permitted by what he termed the
ARPAnet Reference Model (ARM), e.g. optimizing the EBCDIC/ASCII
Presentation of multiplexed Sessions by doing it at the system boundary,
possibly outsourced to the adapter or DCE, where ISORM would forbid
Transport and Presentation functions.)


// William Ricker for The Literary Estate of Michael A Padlipsky