Three ethernet devices on one cable?

[daveshah]

My computing teacher thinks that you can connect three 10/100Base-T devices using the same 2 pairs, with no hubs or switches. Personally, I am skeptical (chances are the devices would not be able to handle the additional loading - as well as a completed impedance mismatch, and Auto-MDIX would go crazy), and think he is confused with the old coaxial Ethernet networks. What are your opinions?

[Bloch]

My computing teacher thinks that you can connect three 10/100Base-T devices using the same 2 pairs, with no hubs or switches. Personally, I am skeptical (chances are the devices would not be able to handle the additional loading - as well as a completed impedance mismatch, and Auto-MDIX would go crazy), and think he is confused with the old coaxial Ethernet networks. What are your opinions?

Well on linux .... http://linuxgazette.net/140/brownss.html

[Monkeh]

My computing teacher thinks that you can connect three 10/100Base-T devices using the same 2 pairs, with no hubs or switches. Personally, I am skeptical (chances are the devices would not be able to handle the additional loading - as well as a completed impedance mismatch, and Auto-MDIX would go crazy), and think he is confused with the old coaxial Ethernet networks. What are your opinions?

Get him to demonstrate it.

(no, it's not going to work)

[tszaboo]

My computing teacher thinks that you can connect three 10/100Base-T devices using the same 2 pairs, with no hubs or switches. Personally, I am skeptical (chances are the devices would not be able to handle the additional loading - as well as a completed impedance mismatch, and Auto-MDIX would go crazy), and think he is confused with the old coaxial Ethernet networks. What are your opinions?

Your teacher is right. Ethernet has collision detection, and retry (the method makes it unusable for realtime applications btw). An Ethernet hub just connects the cables together (with a repeater) so there is your impedance.

mod: Oh damn, I need to learn to read.

[madires]

10Base5 or 10Base2 wiring allows several systems being connected to a single coax cable and to run Ethernet in half-duplex mode (classic CSMA/CD). The physical interface of 10BaseT and 100BaseTX requires two twisted pairs, one to send and another to receive. With a crossover cable two systems could be connected directly. Connecting more than two systems to a TP cable (only 10/100BaseT) is possible but the additional systems would be just wire tapping devices because there can't be an any-to-any communication anymore. For an any-to-any communication a hub is needed, which repeats what it receives to the receive pairs of the connected systems (also half duplex mode).

BTW: There are some passive sniffing adapters for 10/100BaseT using the concept above. For 1000BaseT and higher passive sniffing isn't possible any more due to the physical interface.

[AlfBaz]

As far as installation goes it was a backward step. Run one coax from the server to the next pc, bnc "T" piece and on to the next, BNC terminator on the last pc.
Then UTP came along and it was multiple runs... PITA

[Monkeh]

As far as installation goes it was a backward step. Run one coax from the server to the next pc, bnc "T" piece and on to the next, BNC terminator on the last pc.
Then UTP came along and it was multiple runs... PITA

But at least one faulty cable can't bring the whole system down!

[dfmischler]

As far as installation goes it was a backward step. Run one coax from the server to the next pc, bnc "T" piece and on to the next, BNC terminator on the last pc.
Then UTP came along and it was multiple runs... PITA

But at least one faulty cable can't bring the whole system down!

With hubs it still could (and sometimes did).  The benefits of switches are not just in throughput and security.  They also help a lot in isolating faults to individual cable runs.

Trust me.  Once the ethernet network grows to over a few dozen nodes 10base2 or hub connected 10baseT starts to get a lot more difficult to manage.  Modern switched gigabit ethernet is super easy to manage if you have good switches, so the additional cabling is well worth it.

[AlfBaz]

But at least one faulty cable can't bring the whole system down!

That's a problem for "others"

[Stonent]

Apparently where I work we used to have a guy who took a lot of shortcuts in wiring.

Cat 5 cable with orange/green used for computer and brown/blue for digital phone or orange/green for jack1 and blue/brown for jack 2. Or blue pair for 1 analog phone, brown pair for another analog phone.

True you can do 100baseT over 2 pairs but since most of our computers are gig capable and we have gig switches in some areas, I could see some problems coming up in the future.

And trying to test an outlet that you don't know is wired that way just seems to irritate our fluke etherscope.

[dfmischler]

Cat 5 cable with orange/green used for computer and brown/blue for digital phone or orange/green for jack1 and blue/brown for jack 2. Or blue pair for 1 analog phone, brown pair for another analog phone.

The early 10baseT literature actually promoted the fact that you could do this.  It wasn't until a few years later that the EIA/TIA-568 standard suggested that you really should have multiple cables per user, and that unused pairs in ethernet cables should be reserved for future use.  I remember being involved in wiring a pre-standard building for a company I worked for with 6-pair plenum cable to support multiple services including 10baseT over a single cable per user.

In my current facility a single 4 pair cable is used for each ethernet drop, and other services are very rare because the phones run on POE ethernet.

[Stonent]

Another employer replaced their old token ring system with Ethernet. The cable was Cat4 16mbit capable and they moved to 10baseT. We just got baluns for all the plugs on the walls.

But it seemed to cause a lot of hassle for us lowly people who had to deal with "Domain controller not available" popups at login because they bumped the balun and it partially unseated from the token ring plug.

[NiHaoMike]

With hubs it still could (and sometimes did).  The benefits of switches are not just in throughput and security.  They also help a lot in isolating faults to individual cable runs.

I have seen a bad cable take down a good part of the network at work. Somewhere in the back of one of the test benches, one of the cables connecting two (small) switches had a broken tab. Then someone swapped hardware on the bench and accidentally loosened the connector just enough to stop it working. Without thinking, he ran another feed into the islanded switch. Later, someone else discovered the defective cable and replaced it with a good one, making a loop. Didn't take long for the whole network in that room and part of another room to go down. (I presume the cable started making contact and caused network issues, leading him to inspect the wiring.)

[Whuffo]

My computing teacher thinks that you can connect three 10/100Base-T devices using the same 2 pairs, with no hubs or switches. Personally, I am skeptical (chances are the devices would not be able to handle the additional loading - as well as a completed impedance mismatch, and Auto-MDIX would go crazy), and think he is confused with the old coaxial Ethernet networks. What are your opinions?

Electrically, this could work - you wouldn't be able to use the longest cables, but all three devices should get adequate signal. That said, it's a horrible idea; the packets are routed based on the MAC number of the interface, and network routers, switches, hubs, etc. all depend on each MAC being at the end of a particular Ethernet segment. If you have multiple devices spliced into that segment, than the multiple MAC numbers would produce "undefined" results from the upstream devices.

Really, if you need to connect 3 devices - install a 4 port switch. They're inexpensive enough, and they'll use their internal intelligence to make sure the packets go to the right place.

[kaz911]

With hubs it still could (and sometimes did).  The benefits of switches are not just in throughput and security.  They also help a lot in isolating faults to individual cable runs.

I have seen a bad cable take down a good part of the network at work. Somewhere in the back of one of the test benches, one of the cables connecting two (small) switches had a broken tab. Then someone swapped hardware on the bench and accidentally loosened the connector just enough to stop it working. Without thinking, he ran another feed into the islanded switch. Later, someone else discovered the defective cable and replaced it with a good one, making a loop. Didn't take long for the whole network in that room and part of another room to go down. (I presume the cable started making contact and caused network issues, leading him to inspect the wiring.)

Many decades ago I serviced IT for a financial company in London running token ring network (IBM thing from the past) - and once in a while the WHOLE network would kill itself. We checked and checked and checked - and the intermittent problem stayed hidden. Until one day I was sitting and setting up a new PC - watching their pretty secretary  .... - and she crossed the floor in her super high heels - and kaboom - whole network went down. Someone had pulled a cable under the carpet - and when she stepped on it with her very thin heel - the cable shorted... The moment the heel was lifted - the cable was fine again. So I have made it a point to always watch pretty secretaries. 

[Bored@Work]

If you have multiple devices spliced into that segment, than the multiple MAC numbers would produce "undefined" results from the upstream devices.

No, no, no, no, and no. CSMA/CD would prevent "undefined" results, and devices would just have more entries in their MAC tables pointing to the same interface. Something which isn't unusual at all.

[amyk]

If you Google "passive hub" you'll find lots of schematics that essentially amount to tying the data pairs together, and they do work.

[madires]

As far as installation goes it was a backward step. Run one coax from the server to the next pc, bnc "T" piece and on to the next, BNC terminator on the last pc.
Then UTP came along and it was multiple runs... PITA

The T adapter is used just for 10Base2. 10Base5 required to pierce the cable. The problem with both is that any network change like adding an additional PC causes an outage. And with a lot of systems the wiring becomes a nighmare and you'll see a lot of collisions. TP wiring has fixed all those problems and gives you full duplex and more bandwidth. 

[madires]

I have seen a bad cable take down a good part of the network at work. Somewhere in the back of one of the test benches, one of the cables connecting two (small) switches had a broken tab. Then someone swapped hardware on the bench and accidentally loosened the connector just enough to stop it working. Without thinking, he ran another feed into the islanded switch. Later, someone else discovered the defective cable and replaced it with a good one, making a loop. Didn't take long for the whole network in that room and part of another room to go down. (I presume the cable started making contact and caused network issues, leading him to inspect the wiring.)

A classic one :-) Buy manageable switches with STP (spanning tree protocol) support to supress loop issues.

[madires]

Electrically, this could work - you wouldn't be able to use the longest cables, but all three devices should get adequate signal. That said, it's a horrible idea; the packets are routed based on the MAC number of the interface, and network routers, switches, hubs, etc. all depend on each MAC being at the end of a particular Ethernet segment. If you have multiple devices spliced into that segment, than the multiple MAC numbers would produce "undefined" results from the upstream devices.

How would you connect the send and receive pairs of three 10/100BaseT NICs in a way that each on is able to transfer frames to each other? Ethernet frames aren't routed!!! They are sent/received, repeated (hub) or switched. OSI layer 3 is routed. MACs aren't a problem because each MAC should be unique by design and you can have as much MACs in a broadcast domain as you like or your network elements are able to support.

[PaulAm]

I suppose you could do it, but you wouldn't be able to talk to at least one of the systems.

To directly connect 2 systems with TP and no hub/switch, you need a cross connect cable, an ordinary patch cable will not work.  So your third system could listen to one or the other, but not both.

[nctnico]

Nowadays lots of devices are auto MDX so you don't need a cross cable.

[madires]

I suppose you could do it, but you wouldn't be able to talk to at least one of the systems.

To directly connect 2 systems with TP and no hub/switch, you need a cross connect cable, an ordinary patch cable will not work.  So your third system could listen to one or the other, but not both.

Exactly! One of the three systems would run as a network sniffer but only for one direction. For sniffing also the reverse direction the third system would need a second NIC, i.e. RX pair of NIC #1 conected to the first TP pair and RX of NIC #2 connected to the second TP pair.

[AlfBaz]

As far as installation goes it was a backward step. Run one coax from the server to the next pc, bnc "T" piece and on to the next, BNC terminator on the last pc.
Then UTP came along and it was multiple runs... PITA

The T adapter is used just for 10Base2. 10Base5 required to pierce the cable. The problem with both is that any network change like adding an additional PC causes an outage. And with a lot of systems the wiring becomes a nighmare and you'll see a lot of collisions. TP wiring has fixed all those problems and gives you full duplex and more bandwidth.

[amyk]

I suppose you could do it, but you wouldn't be able to talk to at least one of the systems.

To directly connect 2 systems with TP and no hub/switch, you need a cross connect cable, an ordinary patch cable will not work.  So your third system could listen to one or the other, but not both.

Exactly! One of the three systems would run as a network sniffer but only for one direction. For sniffing also the reverse direction the third system would need a second NIC, i.e. RX pair of NIC #1 conected to the first TP pair and RX of NIC #2 connected to the second TP pair.

Connect RX and TX pairs together and use half-duplex?

[madires]

Exactly! One of the three systems would run as a network sniffer but only for one direction. For sniffing also the reverse direction the third system would need a second NIC, i.e. RX pair of NIC #1 conected to the first TP pair and RX of NIC #2 connected to the second TP pair.

Connect RX and TX pairs together and use half-duplex?

I'm not sure about that. A hub which also implies half duplex mode doesn't send the input received at a port at the same port out again. It just sends the input to all other ports except the one where the input comes from. So we might assume that connecting RX and TX isn't supported by 10/100BaseT. Seems I'm right, in IEEE 802.3-2002 Section 1 about the 10BaseT MAU there are clear statements that a point-to-point connection is supported only between two MAUs, and if more than two DTEs should be connected together a repeater is required. And if I got the description of the Collision Presence function for half duplex mode right, receiving any Manchester code on RX while sending on TX causes a collision. So the answer is: no.