[ih] Ethernet, was Why TCP?

[Larry Sheldon]

On 9/1/2016 08:20, Noel Chiappa wrote:
>     > From: "John Levine"
>
>     > As we all know, Ethernets worked just fine. A lot of people didn't
>     > believe it until they saw it
>
> In defense of the Ethernet sceptics, most Ethernets were operated at traffic
> levels where the collisions which theory predicted at high traffic levels
> weren't a problem.

Which speaks directly to the relative costs issue--low traffic levels is 
almost always (he says unscientifically) what is present at turn-up. 
(and I suspect for all of eternity.)
>
> Liba Svobodova (at MIT at that point) did a bunch of analyses (this would be
> ca. '78 or so) which showed that at low traffic levels, it would be fine, and
> that only at high traffic levels would there be issues. Her analyses, AFAIK,
> were right on target.
>
>
> And of course the _real_ advantage of rings was not so much in the
> token-passing access method, as in the analog lower level they used - all
> simple point-point links. (Great for optical, BTW.)

There was a time when I thought I was going to have to implement token 
ring on a network that consisted largely of 56-kb frame-relay lings in a 
star, with most of the many legs hundreds of miles long.  I wondered 
(but never learned) how long it would take for the token to complete a 
lap in such an arrangement.  (I did think about an earlier life working 
with Teletype™ systems where, when a line was idle, the control station 
would "poll" the network by sending out two-letter address groups.  If 
the addressed station had outbound traffic, it would commence sending 
it. If it did not, it sent the letter "V".  The system could gridlock 
for a number of reasons but the usual one was a station that had traffic 
arriving (on other circuits) faster than it could get rid of 
it--exacerbated by several busy stations on the circuit.  One attack on 
the problem was to poll the busy stations several times in each trip 
around the loop.  I had (have) no idea how to do that with Token Ring™.


>
> Which is why today's 'Ethernet' networks are not (when you lift the covers)
> really CSMA-CD networks: they actually consist of lots of little packet
> switches connected by point-point links. I.e. the current systems use the best
> of both approaches (the simple access mechanism of Ethernet, and the simple
> point-point links of rings).
>
> The only parts of 'Ethernet' left are the packet format, and the host/network
> interface standard. As always, the interface is left when the equipment on
> either side has changed out of all recognition - check out the screw base on a
> mains AC LED bulb...

Reminds me of a question I don't see discussed anywhere--do the 
"Ethernet switches" (that replaced the "hubs" from my day in the sun) in 
fact operate as "multi-port bridges" complete with Spanning Tree, BDUs, 
and the other neat Radia Perlman* stuff?

*Interesting to note (ref. earlier mention of DECnet) she worked for DEC 
when she did that stuff.

-- 
"Everybody is a genius.  But if you judge a fish by
its ability to climb a tree, it will live its whole
life believing that it is stupid."

--Albert Einstein

 From Larry's Cox account.