AUI cables in UK?

From: Paul Koning <pkoning_at_equallogic.com>
Date: Tue Mar 9 09:20:36 2004

>>>>> "Fred" == Fred N van Kempen <waltje_at_pdp11.nl> writes:

>> > Given that this stuff was often hidden in suspended ceilings >
>> and suchlike, it would be quite tricky to check that adjacent >
>> stripes (or two adjacent stripes) were unused (I assume the rule >
>> applies in both directions. Surely it would have been easier > to
>> miss out the unusuable stripes?
 Fred> Correct, this cable was usually on those ceilings, with taps
 Fred> being set up there, and then an AUI dropcable down the wall to
 Fred> the device, or sometimes another converter that ran BNC to
 Fred> small devices.

That would have to be a repeater, since you cannot install side stubs
on an Ethernet coax. That's why thin Ethernet had the BNC T right on
the device. Some installations used wires that look like a single
piece of coax but in fact is a double coax, running from the wall
outlet box to the back of the computer.

I once saw an article describing how to go from thick to thin coax --
it showed a T connector for that, with a terminator! Not good,
terminators never go on the middle of a coax. The correct answer was
simply an N to BNC barrel converter.

>> But I have actually never really understood why. Maybe just to not
>> damage the cable to much with vampire taps? There cannot be an
>> electrical or signal reason for this, as far as I can tell.
 Fred> For signal echo reasons, much like the requiement for
 Fred> terminators on both ends of the cable. If the devices are too
 Fred> close to eachother, there will be too *little* space (read:
 Fred> time) inbetween for them to check for any echo (collision).
 Fred> The transceiver will send out the first bit, and will then wait
 Fred> the minimal time (the time it takes to reach the first possible
 Fred> device on the cable) before it checks for collision. It then
 Fred> sends out the packet (if no coll.) and waits a full timeframe
 Fred> (meaning, the time maximally needed to reach a terminator, and
 Fred> then back) for a collision. If none, we have sent a frame.
 Fred> Otherwise, we have a collision.

That's close but not quite.

The spacing rule is to avoid having multiple taps that are integer
multiples of a wavelength apart. Each tap unavoidably creates some
disturbance of the cable impedance, which means electrical
reflections. Those have to be kept low enough or the collision detect
machinery breaks (it will mistake an excessive reflection for a
collision).

The maximum round trip delay on an Ethernet is 512 bit times, or 64
byte times, which is why the minimum packet size is 64 bytes. That
ensures that when two stations transmit overlapping packets, both will
see the collision. If shorter packets were used, it would be
possible for packets to overlap at a station in the middle, yet the
senders don't see it as a collision. Conversely, a "fragment"
generated by a collision is always smaller than 64 bytes.

There is no such thing as "send the first bit and wait to see if there
is a collision". Instead, collision detection works by observing the
voltage on the cable and comparing it with what you would get from a
single transmitter. As the Ethernet spec puts it, "...the average DC
signal on the coaxial cable implies a simultaneous transmit attempt by
more than one station (a collision)..." (7.4.2.3.)

     paul
Received on Tue Mar 09 2004 - 09:20:36 GMT