Path: utzoo!attcan!uunet!aplcen!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!swrinde!ucsd!ucbvax!GATEWAY.MITRE.ORG!barns
From: barns@GATEWAY.MITRE.ORG
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Mourning of the passing of the ARPANET
Message-ID: <9006271300.AA25518@gateway.mitre.org>
Date: 27 Jun 90 13:00:01 GMT
References: <1990Jun26.190232.8554@Solbourne.COM>
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 52

I'll let someone else talk about net 89.  However, here is some history
on network numbers.

When IP was first conjured up, there was not a notion of class A/B/C
addresses.  At that time it was not generally thought that there would
be a huge number of networks, but rather a smaller number of possibly
large networks.  The 32 bit IP address was 8 bits Network, 24 bits Host.
"They" thought it would be a good idea to preallocate the network
numbers to all the significant networks (those that some Internet
participant might presumably join).  I seem to remember that there were
on the order of 20-30 networks identified and preassigned.  These
included TELENET, TYMNET, DATAPAC, etc.  There was a list and it can
probably be found in some ancient RFC.

A little while later "they" decided that preallocation was not such a
good idea and instead they would give out network numbers only to networks
that had EGP routing glue to The Internet.  Some of the already assigned
network numbers were deassigned because there was no connectivity and no
specific plan to achieve it.  It was also around this time that
the threat (or promise, depending on your point of view) of campus
nets, building nets, and random Ethernets became clearer, so classes
A, B, and C were invented, and the high order bit coding trick was used
to provide compatibility with the previously assigned network numbers
that were actually in use.  So there were a lot of holes in class A and
most of the new demand was in class C, for the random Ethernets.

Obviously the people who already had working networks didn't want to
change their addresses.  This is the main reason why some nets that don't
seem to be big enough to warrant class A addresses have them anyway.
Less obvious but also an issue - some sites had gateways that could not
conveniently be gotten to cope with the revised addressing plan.  I
think that Stanford was one such, but I don't remember the exact problem.
Bear in mind that gateways weren't off-the-shelf in those days.  (Neither
were hosts, sometimes.  Especially the networking interface hardware and
software.)

Subnets are more recent than all of these things, and multicast (class D)
addresses are yet more recent.  Class B became popular with the advent
of subnetting.

We probably don't have much call for class A network numbers nowadays,
but I can think of at least one possible source of demand.  When a
logical network uses end-to-end encryption devices, there are some
extra demands on the address space, such as wanting to be able to
address some sub-processors in the encryption box, and wanting to carry
the encryption domain identification in a subfield of the address.  To
put all of this (plus the host id) into 16 bits with fixed size subfields
would cramp the fields more than is nice.  Even 24 bits isn't vast, but
it's better.  DOD has net 21 for this purpose already; I don't know
whether others are or will be needed, but it's conceivable that they might.

Bill Barns / MITRE-Washington / barns@gateway.mitre.org