[ih] Internet-history Digest, Vol 2, Issue 16

Noel Chiappa jnc at mercury.lcs.mit.edu
Wed Nov 6 07:29:51 PST 2019

    > From: Clem Cole

    > You need to agree on what an '*internet*' is ...  I'm not sure what the
    > right answer is here

_An_ internet I would say is a collection of networks which are prepared for
forward packets from a host on one to a host on another, based solely on the
address in the bottom protocol layer above the 'local' network headers - which
means there must be a coherent system-wide namespace for those addresses.

One can argue as to whether or not those packets are reliable or unreliable,
etc, and similar issues, but the above definition cannot be stretched too far,
otherwise things like 'a collection of X.25 networks using X.75 to create
connections from one host to another' becomes an 'internet'. Just being able to
get data from one network to another doesn't make the resulting confection an

So PUP, CHAOS, TCP/IP etc were all 'internet' protocols. So the date of the
'first internet' packet would be different for 'first PUP internet' packet,

    > I think that somehow that idea of packets flowing both ways needs to be
    > in the definition of a full 'internet.'

Yes, but access controls may limit it.

    > I know that CMU and MIT built something internally to connect local
    > hosts and allow them to connect to the directly connected ARPANet hosts
    > that had IMP connections. ... IIRC MIT used ChaosNet protocols.

Yes, see A.I. Memo 628, 'Chaosnet', Section 5.9, "Arpanet Gateway".

Basically, the 'ARPA' CHAOS service connected a CHAOS reliable stream to an
NCP connection; the destination host and port were specificied in the CHAOS
connection open. So on the plus side, it could be used for any NCP
application, but on the downside, you had to know the CHAOS address of a
machine that provided that service, and the NCP address of the ultimate
destination, and generally the application on the 'source' host had to be
modified to know about the ARPA service to be able to use it.

    > Also for the sake of argument, what about the UUCP network?  It did not
    > support remote login, but it did support remote file transfer, email and
    > remote execution of jobs? ...  - does that count as an internetwork?

I would say 'no', because it didn't provide the ability to carry an individual
packet from one host to another one anywhere. The details of UUCP are now gone
from my mind, but ISTR that it provided reliable streams?

Internet-history mailing list
Internet-history at elists.isoc.org

More information about the Internet-history mailing list