From: Joe Touch <touch(_at_)ISI(_dot_)EDU>
...
At IP, as Bob Braden said.
SMTP is _over_ IP.
Multicast _redefines_ IP (or portions of the address space thereof); it
could be argued that a service provider sells 'Internet' without selling
multicast IP.
That grossly overstates the difference between multicast IP services and
classic IP services. For one thing, many multicast applications work
...
'Internet' is about speaking IP and ICMP.
And which multicast application isn't?
There are many variants of routing; none are required to be deployed
_throughout_ the Internet. Static routes are sufficient, and 'who speaks
what routing protocol' and 'what the routes mean' (CIDR included) is a
matter of consensus among parties exchanging a single routing protocol,
not an Internet-wide requirement.
What is the relevance of that to the question of when a package labeled
"The Internet" is a fraud?
BGP4 is a lot newer than multicasting. Would you say that an ISP with
broken routing that makes a significant part of the net unreachable is
still legitimately selling "The Internet"?
It's also a of a stretch to call the 1985 change of class D from
"unused" or "reserved" to the multicast space a redefinition of the
IP address space.
Under classic IP, class D was defined as unused/reserved;
under multicast IP, class D is now defined as multicast.
That is the purest form of the change of a definition.
While it affects only a portion of all IP packets, it did redefine
the meaning of that portion.
In exactly what way did it redefine any IP header bits? Yes, what routers
must do and the link-layer destination of some IP addresses was tweaked,
but little more than what RFC 1122 did to broadcast addresses. The
multicast changes were no than many other changes since 1085, including
new ICMP types. Would you say an ISP selling "Internet email and web
hosting" is honest if its hosts don't do slow start, because slow start
wasn't there at the beginning? What about the many other new requirements,
such as egress filtering and not advertising routes to RFC 1918 networks?
Or the deprecation of RFC 822 promiscuous relaying?
(it redefined the meaning of values of the space, not the partitioning
of the space).
More than the meanings of other large chunks of the 32-bit IPv4 space
listed in RFC 960 as "reserved"? I'll grant the word "reserved" was a
little general, but not enough to matter.
The question is not "what was NFSNet in 1985." It is, "in 2000, at what
point is a vendor claiming to be selling "The Internet" guilty of false
and misleading advertising?" Would you really allow a vendor to sell
exactly those services that were available in 1985? Why not pick 1970?
If you must do it by date, 1990 or 1995 makes a lot more, but still not
much sense. "The Internet" is not a static thing.
When (and if) IPv6 takes off, would you say an "Internet" package could
exclude the IPv6 universe, including hosts reachable with an embedded IPv4
address? If you're consistent, your answer must be "yse."
If it were relevant, I'd ask about the cause for your unreasoning prejudice
against multicasting. You might be able to support a claim that
multicasting in the Internet is a bad idea, can't work in many ve large
internets (small 'i'), is useless except for either trivial applications
or applications that can do as well or better with broadcasting (e.g.
NTP), or similar. Claiming that multicasting is not part of The Internet
because it wasn't in TIP's and IMP's in 1970 is something else.
Vernon Schryver vjs(_at_)rhyolite(_dot_)com