[ih] early networking

[Jack Haverty]

On 4/20/24 10:16, Matt Mathis via Internet-history wrote:
> In my mind the crucial event was to split TCP and IP into
> separate protocols, such that there was deep architectural enforcement of
> the hourglass and the orthogonality of the upper and lower protocol
> layers.   This orthogonality means that the cost of  maintaining M
> applications over N link types scales as O(M)+O(N).

Agreed.

FYI, this notion of an "hourglass" with IP at the neck was foreshadowed 
earlier in the design of the basic Arpanet protocols. E.g., the notion 
of a "Network Virtual Terminal" or NVT was the "neck of the hourglass" 
which simplified the MxN problem for connecting all sorts of terminals 
(each with their own characteristics) to all sorts of computers (each 
with its own idea of what a terminal looked like).   The NVT allowed the 
two groups to work independently to make their components "look like" an 
NVT to the other side of the neck.

That "hourglass" technique became an essential principle of network 
design.   In the case of terminals, it was a way of interconnecting two 
often disparate worlds, so that, for example, an IBM terminal speaking 
EBCDIC could interact successfully with a TOPS-20 mainframe expecting to 
see terminals using ASCII and RS-232.   The concept was also used 
elsewhere, e.g., to define various kinds of "network file systems" such 
as NFS.

See, for example, https://www.rfc-editor.org/rfc/rfc318

The introduction of a "neck" also facilitated the creation of other 
"hourglasses", if you could define a way to convert as needed to be 
compatible with the neck's constraints.   In the case of TCP/IP, that 
enabled the creation of UDP, running directly over IP, as a hopefully 
more appropriate way to carry voice traffic, for which timely delivery 
of as much as possible was more important than getting everything 
delivered.

That diversity also motivated the definition of TOS (Type Of Service), 
anticipating that the underlying IP service might need to handle 
different kinds of traffic in different ways - as soon as someone 
figured out how to do so and there was enough CPU and memory to 
implement it, and metrics within the Internet evolved to use actual 
transit time instead of "hops".

Jack

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 665 bytes
Desc: OpenPGP digital signature
URL: <http://elists.isoc.org/pipermail/internet-history/attachments/20240420/394808ce/attachment.asc>