From: V Bruce Hunt
Sent: Tuesday, June 6, 2023 1:02 PM

Hi,  I can confirm that the Ariel project (later Z-Net) at Zilog was the basis
of Local-talk.  The story is roughly the following.  I got a call from Steve
Jobs after giving a presentation and Bob and John's distributed computing
seminar at Stanford asking me to come and talk to Apple.  I met with Steve Jobs
and another person whose name was (I think) Steve Levy .  They asked me to
explain how the Ariel network worked.  I spent an afternoon over on Bandley
drive (not all that far from Zilog's offices on Bubb Road) with Jobs and Levy.
Both Ben Laws (my co-conspirator and hardware genius on the Ariel project) and I
worked for Joe Kennedy and Joe reported to Charlie Bass.  I don't remember why
Ben could not join me.  In any event I explained that we used the SIO's built-in
SDLC/HDLC protocol using an NRZI encoding.  The reason for this is that
SDLC/HDLC uses a technique called bit-stuffing in which after every 5 bits one
bits a zero bit is inserted between the start and end packet indicator ( whose
encoding is 0b01111110) so using the NRZI encoding guaranteed that a clock edge
would appear at least every 6 bits.  It also provided elegant packet framing.
It was then easy to extract the clock out of the data by the use of the NRZI
encoding.  I think we used line edge detection for collision detection.  The
idea was that if we were about to drive the data line  up or down and it is
already down or up then clearly we have either a collision or stuck at
condition.  That is, we only looked for collisions at data transitions.  These
techniques worked really well in practice and it was incredibly simple to
implement in hardware.  As a result of the conversation Apple adopted the Zilog
SIO for both the serial port and what became local-net.

A few years later (late '83 I think)  Steve Jobs called me up and asked me to
consult on a performance problem that Apple was having with Local-talk.  This
was a couple of months before the Macintosh was introduced so late '83.  The
problem was that on the 230Kbit/sec signaling rate they were not getting
anywhere close to that bandwidth on the network.  This turned out to be a
problem with the back off timing and the in-elegant collision detection method.
I showed them how to improve the performance with no hardware changes to get
them to about 80% of the available bandwidth; and with the addition of a single
chip how to get it to exceed 90% under heavy load.  Steve opted for the 80%
because as he said, Apple had a large number of Mac's already built getting
ready for the launch.  This was after Fouad Tobagi and I published our paper on
the performance of collision detection networks and I had done some work on
optimal back-off algorithms on random-access networks with goals of stability,
maximizing throughput and minimizing average delay.