[ih] Early History of the Internet

[Karl Auerbach]

You are right about complexity of hardware.  Ethernet was genius - a 
marvel of simplicity in the way it extrapolated from the way we humans 
talk in a noisy room to the world of electronic communications with 
minimal inter-device coordination.  Was the Aloha net the first to do 
this kind of thing?

But lest we think Ethernet went smoothly at first, let us pause and 
remember the original 3COM 3C501 Ethernet card - it took too long to 
recover from a transmit operation to be ready to receive a response 
packet from a fast peer.  I think I banged my head against several walls 
trying to get a 3C501 work on a slow PC/AT using Romkey/Bridgham PC/IP 
when the other end was an early Sun workstation.

DCS did indeed depend on the properties of their network - it was 
essentially a broadcast medium with an additional interesting function: 
When the packet went around the ring and came back to the sender there 
was a flag that indicated whether any device had accepted the message 
(and copied it from the ring.)  This served as a kind of "someone is 
interested" acknowledgement with effectively zero cost.

(I vaguely remember another similar flag that was set as the packet went 
around the ring, but my memory has faded about that.)

If I remember correctly the actual transmission signal format was based 
on telco T-1.

I get the impression that Dave Farber is rather pleased that his project 
undermined the IBM patent on token rings.

         --karl--


On 1/10/24 12:20 PM, Dave Crocker via Internet-history wrote:
> On 1/10/2024 12:03 PM, Karl Auerbach via Internet-history wrote:
>> I first became aware of that work in the rather early 1970s when I 
>> was at SDC.  Frank Heinrich (one of Farber's students) had worked on 
>> a distributed file system for DCS.
>
> When Dave showed me one of the network cards for DCS, the 
> token-vs-contention battles were underway.  He explained some aspects 
> of the card, including the part that did token resolution at startup.
>
> It took a quarter of the real-estate and used a contention-based 
> scheme.  Given this, it seemed clear to me that, absent extremely 
> strong needs for highly predictable access times, this meant that 
> Ethernet would win handily.
>
> d/
>