[ih] Early interconnection between Ethernets and Arpanet

[Craig Milo Rogers]

On 10.03.11, Matthias Bärwolff wrote:
> is there any documentation (or tacit knowledge on this list) on early
> Ethernet-Arpanet gateways (obviously other than those built using IP in
> the late 1970s)? I found some documentation on how Alohanet terminals
> connected to the Arpanet, and how the NPL guys in the UK connected their
> IBM machines to their TIP, plus some notes on how people struggled to
> get the SRI port expanders working; but nothing on the much more obvious
> candidate, Ethernet, given its early popularity.

	This may not be early enough for you, but here is my
contribution, from memory.  I started working at USC/ISI on
7-Jul-1980.  My first major assignment was to interface a Xerox laser
printer, called the Penguin, with the Tenex/TOPS-20 ARPANET hosts and
other systems running at ISI, using an available PDP-11 system as a
gateway.  The Penguin printer connected to a 3-MBit Ethernet and spoke
a PUP-based file transfer protocol, EFTP; it also offered a printer
status protocol, I think.  The Tenex/TOPS-20 systems connected to the
ARPANET and offered experimental IP services.  The PDP-11 ran the EPOS
operating system, a real-time operating system developed in-house at
ISI by Steve Casner and others.  It had connections to the 3-Mbit
Ethernet and to one of ISI's ARPANET IMPs.

	The Ethernet/Arpanet gateway I constructed offered the
following categories of services:

1)	Printer connections, implemented by dynamically translating
	TFTP file transfers (over UDP over IP), from the Tenex/TOPS-20
	systems, into EFTP (over PUP) to the Penguin.  It was convenient
	that the protocols involved were similar enough that translations
	could be done on a per-packet basis, with no byte stream repacking
	required in the gateway.

2)	General IP gatewaying services: various systems at ISI, including
	PDP-11's running EPOS and Three Rivers PERQ systems, used the
	Ethernet-to-ARPANET gateway as their primary path to the outside
	world.

3)	Diagnostic and experimental services, such as UDP Echo, UDP sink,
	traffic statistics, and so on.

	An instance of this gateway system was installed at DARPA, to
service a Penguin printer installed there.  There was a problem
installing the gateway, which turned out to be a ground loop in the
distant host interface to the IMP.  Once that was solved, the system
required minimal support during its lifetime.

	My gateway software went through two major iterations.  When I
was introduced to EPOS, Steve Casner said that EPOS IPC was very
cheap, and I should not be shy of using it.  So, I implemented the
Ethernet manager, IP manager, UDP manager, TFTP manager, PUP manager,
EFTP manager, and print connection translater all as seperate
processes. I demonstrated that EPOS IPC was not sufficiently cheap for
this system design.  The second implementation minimized IPC usage,
and achieved much better packet processing rates.

	I assume you;ve already read the documentation on Dave Mill's
Fuzzball systems.

					Craig Milo Rogers