[TUHS] BerkNet

[Erik E. Fair]

	Date: Fri, 21 Nov 2014 01:07:59 -0500 (EST)
	From: Cory Smelosky <b4 at gewt.net>

	I take it you actually understand BERKNET's addressing then? I could 
	never figure it out!

BerkNET was its own thing: it was effectively a store & forward batch/file
based networking system which could handle E-mail, print jobs, and remote
execution of commands for your account on another machine, if you sent the
password to the account along. It's somewhat similar to UUCP, actually. Eric
Schmidt (of Sun, Novell, & Google) wrote the code as a UCB grad student,
because (as I heard it) the Berkeley Computer Center wanted to be paid money
every time an operator had to hang a magtape on a tape drive, and the CS
department was tired of being bled to move small files around.

BerkNet used one single ASCII character designation per machine, originally
just 26 allowed, and later extended to numerals for a total of 36 (before
Ethernet & TCP/IP obsoleted it). The routing table for BerkNet was a
statically-compiled table in every instance of its primary daemon, and if the
network topology changed sufficiently, each daemon had to be modified and
recompiled. No redundancy allowed in the network, if I recall correctly.

Pretty slow convergence for changes to the routing table.

Longer names were also allowed as aliases for the single letter, and that's
what was sent to the rest of the world in E-mail addresses, e.g.
cory:cc-54 at berkeley (Computer Club account #54, on the Cory Hall PDP-11/70;
I think Cory's letter was "y") was my first ARPANET-reachable E-mail address.
There might be some instances of that in the HUMAN-NETS or SF-LOVERS archives
from 1981. I suspect that BerkNet's colon separator for host:file was how
the rcp command got that syntax, and probably how ssh inherited it.

Google turned up the following: http://typewritten.org/Articles/berk-net.html

Being RS-232 serial-based, the interrupt loading was horrific ... so they
restricted the bandwidth to 1200 baud (plus, there were some rather long RS-232
cable runs between buildings), unless you used a serial interface card that
had some input buffering & DMA I/O like the DH-11. The DZ-11 was
contraindicated. To further lower overhead, there's even a TTY line discipline
for BerkNet, so you don't wake up the daemon until a full packet arrives, even
though the TTY interface is set to "raw" mode.

I was for a time a system administrator for the "x" machine at UCB: the Onyx
Z8002 installed for the undergrads in the basement of Evans Hall (room B50).
That's also the machine on which "B news" was written by Matt Glickman.

Later, I ran a small BerkNet (before we got Ethernet) at Dual Systems in
Berkeley, between the Dual 83/80 mc68000-based S-100 systems in the various
departments (engineering, sales, manufacturing/test), before we got an S-100
Ethernet card working and ran thick Ethernet. We were able to run it at 19,200
baud because Dual made some really sweet, 256-byte input buffer, DMA I/O,
4-port serial cards: the SIO-4/DMA, based around the Zilog z8530 DUART.

I insisted that we wait for ARP to be done before we deployed Ethernet &
TCP/IP, because evil old hack of grabbing a Class A IP network number,
pretending that the first three MAC address bytes were always the same (after
all, everyone always uses Ethernet interfaces from the exact same manufacturer
in every host on a given LAN, right?) and mapping the last three MAC bytes
into the host part of the Class A wasn't going to fly in the real world.

The hacks we used to do to make these turkeys fly ...

	Erik <fair at netbsd.org>