Ethernet without hubs

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

I know Ethernet is a shared media protocol, for example the original 10Base2
"bus" LAN's, but all modern implementations of ethernet LAN's use
twisted pair cabling, and hubs/switches creating a star topology
network

My question is: Is it possible to use twisted pair cabling to create
an ethernet LAN without a hub? (of more than 2 devices, not just using
Xover cable)

Theoretically, the ethernet protocol can do this, but I'm wondering if
there are limitations in the network adaptors that would stop this
working. I'm asking as I'm thinking of using ethernet to connect some
embedded devices, and being able to daisy-chain them using twisted
pair would be ideal

Slightly obscure question, any theories welcome, especially if someone
wants to massacre some network cards and cable and actually do it

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Alex Holt wrote:

> I know Ethernet is a shared media protocol, for example the original 10Base2
> "bus" LAN's, but all modern implementations of ethernet LAN's use
> twisted pair cabling, and hubs/switches creating a star topology
> network

> My question is: Is it possible to use twisted pair cabling to create
> an ethernet LAN without a hub? (of more than 2 devices, not just using
> Xover cable)
>
> Theoretically, the ethernet protocol can do this,

(snip)

The theory of UTP ethernet is different than of coaxial ethernet.

It would be possible with the appropriate transceiver and for
much more limited distances than coax, to run a shared ethernet
over a single pair of a twisted pair cable. The resistance
and impedance tolerances would be much tighter than 10baseT.

Coaxial ethernet collision detect is very different from
10baseT, and it is that difference that prohibits what you
want to do.

-- glen

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

"Alex Holt" <alex@apholt.net> wrote:

> I know Ethernet is a shared media protocol, for example the original
10Base2
> "bus" LAN's, but all modern implementations of ethernet LAN's use
> twisted pair cabling, and hubs/switches creating a star topology
> network

And the twisted pair approach is electrically different from the
10BASE-2 bus. In the coax system, each host is connected to the shared
medium through a single transceiver, much as a common channel wireless
would be. Until 1000BASE-T came along, all the twisted pair standards
separated the transmit and the receive pairs. Which meant that there had
to be an intervening box that switched any transmit pair from any of the
Ethernet's hosts over to the receive pairs of all other LAN hosts. (As
you have indicated, if you only have two hosts in the LAN, a cross-over
cable can indeed be used to eliminate the intervening hub. But this
trick doesn't work for more than two hosts.)

> My question is: Is it possible to use twisted pair cabling to create
> an ethernet LAN without a hub? (of more than 2 devices, not just using
> Xover cable)
>
> Theoretically, the ethernet protocol can do this, but I'm wondering if
> there are limitations in the network adaptors that would stop this
> working. I'm asking as I'm thinking of using ethernet to connect some
> embedded devices, and being able to daisy-chain them using twisted
> pair would be ideal

So it follows from the above that for 10 and 100 Mb/s Ethernets, it's
not so easy to create a bus system with twisted pair cabling. But maybe
one possibility would be to design something that works like 10BASE-2 to
run over twisted pair. Each transmitter would be a current source, so
that a transmitting host would not create a low impedance drain on
another host attempting to transmit at the same time. A collision would
be detected by a higher than normal voltage on the line, as collisions
are detected in 10BASE-2.

I think the biggest problem with this approach is that cable impedance
must be very predictable and very uniform throughout, or you'll end up
with false collisions or undetected collisions and everything wil fall
apart. That's an advantage of coax over twisted pair. But regardless,
this is *not* 10BASE-T, so it won't work with 10BASE-T or 100BASE-TX
NICs. 10BASE-T and 100BASE-TX NICs expect the transmit and receive
functions to be conducted on their own, separate copper pairs.

I suppose you could use an entirely different approach, like RS-485.

But what about 1000BASE-T? 1000BASE-T operates over 4 *bidirectional*
twisted pair interfaces, using a 5-level AM scheme for each pulse (the
equivalent of two bits of data transmitted over each pulse, over each of
the twisted pairs). Reading through Clause 40 of IEEE 802.3, it would
seem that at some superficial level, a shared bus might actually be
possible. But the system is a master-slave design, with a lot of
handhaking going on between master and slave, and the interface is never
silent. So again, trying to force-fit this into a shared bus among
multiple hosts is not going to be straightforward.

Still, it's an intriguing question, now that 1000BASE-T has changed the
way the twisted pair media in Ethernet work.

Bert

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <Hwr3K8.6qp@news.boeing.com>,
"Albert Manfredi" <albert.e.manfredi@nospam.com> wrote:

> But maybe
> one possibility would be to design something that works like 10BASE-2 to
> run over twisted pair. Each transmitter would be a current source, so
> that a transmitting host would not create a low impedance drain on
> another host attempting to transmit at the same time. A collision would
> be detected by a higher than normal voltage on the line, as collisions
> are detected in 10BASE-2.
>

This is precisely how the Tut Systems Ethernet line extenders work. They
use 10BASE-2 signaling (there is a standard 10BASE-2 transceiver chip at
each end) coupled through a balun (balanced/unbalanced line transformer)
to a single twisted pair. Now, they aren't trying to make a multipoint
system, but conceptually it could be used that way.

> I think the biggest problem with this approach is that cable impedance
> must be very predictable and very uniform throughout, or you'll end up
> with false collisions or undetected collisions and everything wil fall
> apart.

That's important, but it turns out that the toughest problem is getting
the signal to be "perfectly" balanced when driven onto the twisted pair.
Any imbalance becomes common-mode on the pair, which can easily radiate
well beyond legal limits. The "magic" in the Tut product is a patented
balun that provides about 60 db of common-mode rejection; that's a
1,000,000:1 ratio.


>
> But what about 1000BASE-T? 1000BASE-T operates over 4 *bidirectional*
> twisted pair interfaces, using a 5-level AM scheme for each pulse (the
> equivalent of two bits of data transmitted over each pulse, over each of
> the twisted pairs). Reading through Clause 40 of IEEE 802.3, it would
> seem that at some superficial level, a shared bus might actually be
> possible.

Not really. While a given pair is used for both transmit and receive,
they are still "logically" separate channels; great pains are taken in
the transceiver to separate out the two sets of signals. Collision
detection is still performed in the same logical manner as for
10/100BASE-T, not like the "current combining" method of 10BASE-2.

>
> Still, it's an intriguing question, now that 1000BASE-T has changed the
> way the twisted pair media in Ethernet work.
>

It hasn't changed the way twisted-pair media work for Ethernet. The only
reason for combining transmit/receive onto the same pairs was to avoid
requiring *eight* pairs (four in each direction), or having to use an
even more complex coding scheme to get 1 Gb/s onto only two pairs.


--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 395-1966 FAX

Send replies to: usenet at richseifert dot com