> On Jun 27, 2018, at 1:45 PM, Eric Smith <space...@gmail.com> wrote:
>
> On Wed, Jun 27, 2018 at 10:59 AM, Paul Koning <paulkon...@comcast.net> wrote:
> > On Jun 27, 2018, at 12:36 PM, Eric Smith via cctalk <cctalk@classiccmp.org>
> > wrote:
> > Collision detection was the reason (or at least _a_ reason) why the spacing
> > of taps on the 10BASE-5 "thick" Ethernet cable was required to be an exact
> > multiple of 2.5m. It was never clear to me why this was not also a
> > requirement for 10BASE-2 "thin" Ethernet.
>
> Yes, to avoid false alarms. The purpose of the spacing rule is to ensure
> that there is enough signal integrity that you do not get spurious collision
> indications due to reflections off the impedance variations along the cable.
> On a segment with few transceivers, there is enough margin that the rule
> doesn't matter. This is why 10Base-2 doesn't have that rule: the station
> count limit is low enough that it isn't needed.
>
> Interesting! I won't disagree with what you're saying, since I'm ignorant of
> these details, but in my experience 10BASE-2 networks usually had far _more_
> nodes on a network than any 10BASE-5 network I saw. I routinely saw over 100
> nodes on a 10BASE-2, but I never saw more than 20 or so on a 10BASE-5. (There
> certainly may have been larger 10BASE-5 networks; I only ever saw about a
> dozen 10BASE-5 networks.)
I can believe that. But the limit for 10Base-5 is 500 meter segment length,
100 transceivers. For 10Base-2, the limits are 185 meters and 30 stations.
Note that's per segment; the whole network can be larger when repeaters are
used. But the segment limits are the ones that relate to the electrical and
signal integrity aspects of collision detect. I hope your 100-node thinwire
network wasn't a single segment; such a config would run only if you're very
lucky.
The first Ethernets I saw were at DEC, in engineering buildings. They had long
runs of 10Base-5 coax going all over the building, with transceivers for every
node and a whole lot of nodes total. Those were probably pushing the limits.
This reminds me of a research project done at DEC that at one point was
discussed as a possible product but didn't happen: an Ethernet segment mapping
device. It was called "packet voltmeter". The idea was that you'd have one at
each end of the cable (it replaced the terminator). It would build a table of
source addresses and packet signal amplitude. You could then combine the
measurements at the two endpoints, plus the known cable attenuation, to make a
physical map (with tap placement) of each Ethernet node.
The arrival of thinwire with its short segments and star wiring, and especially
twisted pair with point to point wiring, made this idea not suitable for a
product. But the concept was pretty amazing, and it did work.
paul
