[ih] History from 1960s to 2025 (ARPANET to TCP)

[John Gilmore]

Jack Haverty via Internet-history <internet-history at elists.isoc.org> wrote:
> So, my basic question for History is "Why did the architecture
> change?" Were the arguments for a separate network switch (e.g., an
> IMP) no longer applicable?  Did the technology explosion during the
> 70s have some effect?  What was the reasoning behind the decision to
> move the "virtual circuit" mechanisms from the network (IMPs) to the
> hosts?

A major change was from IMPs guaranteeing to their hosts that they would
deliver each packet.  The replacement, IP, *never* promised to deliver a
packet; it was up to the host to re-send the packet if it cared for
reliable delivery.  Thus the network did not offer virtual circuits,
and the host had to synthesize one (with TCP) if the host wanted one.

As RFC 791 (Internet Protocol, by Jon Postel) said:

  The internet protocol does not provide a reliable communication
  facility.  There are no acknowledgments either end-to-end or
  hop-by-hop.  There is no error control for data, only a header
  checksum.  There are no retransmissions.  There is no flow control.

This architecture change also makes it possible to build the core of the
network out of simpler, faster, cheaper IP routers, because the network
only depends on them to do a few things (addressing, routing and
fragmentation).

I had always heard that the reason for this change was that one day the
ARPANET froze up because all the IMPs' buffers were full of packets that
they could not discard (because they had promised to deliver them), so
they could not process any new incoming messages.  The cure for this, at
a protocol level, was to let routers freely discard packets that were in
their buffers, whenever convenient or necessary.  That required that the
hosts be prepared to check acknowledgements and retransmit if needed.

That's my memory -- but does the history support it?  Not so far.  In
looking today, I found three RFCs and another few items that covered
ARPANET-wide crashes, and none of them pointed out this particular
failure mode.  See:

  "Software Checksumming in the IMP and Network Reliability"
  J. McQuillan, BBN; 1973-06-20
  https://www.rfc-editor.org/rfc/rfc528.txt

  "Christmas Day Lockup - Harvard IMP hardware problem leads it to
  broadcast zero-length hops to any ARPANET destination, causing all
  other IMPs to send their traffic to Harvard (25 December)" 1973
  https://www.cs.kent.edu/~rothstei/10051/history/Hobbes-Internet-Timeline.html

  "On a Possible Lockup Condition in the IMP Subnet Due to Message Sequencing"
  L. Kleinrock, H. Opderbeck, UCLA; 1974-03-00
  https://datatracker.ietf.org/doc/html/rfc626

    (This mentions the Christmas lockup, and cites to: BBN Report
    No. 2717, "Interface Message Processors for the ARPA Computer
    Network," Quarterly Technical Report No. 4, January 1974, which
    might have contained a less terse description of the Christmas
    lockup (which Kleinrock says was not on Christmas Day, but on
    1973-12-21).  I have not found an online source of this BBN report
    yet.)

  "ARPANET Lessons"
  Leonard Kleinrock; 1976 International Conference on Communications
  https://www.lk.cs.ucla.edu/data/files/Kleinrock/arpanet_lessons.pdf

    Detailed description of the "Christmas Lockup" and other forms of
    network deadlock.

  "Vulnerabilities of Network Control Protocols: An Example"
  Eric C. Rosen, BBN; 1981-07-00
  https://www.rfc-editor.org/rfc/rfc789.html

Can anyone point us to better sources for this architectural change?

	John