Ethernet Concepts and Questions

[Umesh Shetty]

Hi All,

I've few questions on how ethernet operates under different circumstances and I hope to get some really good answers from all you people here.

• Fastethernet at full duplex uses 2 pairs or the wire RX (wire 1 and 2) and TX (wires 3 and 6) . Incase the Fastethernet port is set/negotiated to half duplex will it continue to use only 1 pair to both send and receive and hence needing to do CSMA/CD to avoid collissions.

• A gigabit ethernet uses all 4 pairs of the wire if the Gig port is negotiated to 100 Mb speed will it start using only 2 pairs instead of 2 ?

• Many Gig etherne ports are capable of 10/100/100 interms of speed and half and full interms of duplex. So is the half duplex possible only when the speed is 10 or 100. Or with a speed of 1000 also half duplex can be acheived ? If yes how many pairs of the wire will it use ?

• I case of Ethernet the speed negotiation happens by the exchange of FLP's which are electrical signals. What is it in case of fibre media ?

Hope to get some gr8 answers

Thanks in Advance

Regards

Umesh

[Peter Paluch · Accepted solution]

Hi Umesh,

Fastethernet at full duplex uses 2 pairs  or the wire RX (wire 1 and 2) and TX (wires 3 and 6) . Incase the  Fastethernet port is set/negotiated to half duplex will it continue to  use only 1 pair to both send and receive and hence needing to do CSMA/CD  to avoid collissions.

This would not be a correct assessment of the situation. FastEthernet always uses 2 pairs, one for Rx, the other for Tx. The duplex setting has no influence on what pairs are used. On a single TP cable, there is in fact no way of creating an electrical collision. However, if there was a hub, two stations operating in full duplex would cause the collision in the hub's circuitry. That is why even with half-duplex and 2 distinct pairs for Rx/Tx, Ethernet has to perform CSMA/CD to avoid collisions.

A gigabit ethernet uses all 4 pairs of the wire if the Gig port is  negotiated to 100 Mb speed will it start using only 2 pairs instead of 2  ?

You mean to say "2 pairs instead of 4". Yes, that is correct. If a Gigabit NIC negotiates 1Gbps, it will use all 4 pairs. If it negotiates 100Mbps or 10Mbps, it will use 2 pairs.

Many Gig etherne ports are capable of 10/100/100 interms of speed  and half and full interms of duplex. So is the half duplex possible only  when the speed is 10 or 100. Or with a speed of 1000 also half duplex  can be acheived ? If yes how many pairs of the wire will it use ?

The half duplex is a very rare thing with 1Gbps Ethernet. Theoretically, it is possible to operate in half duplex if you are using 1Gbps hubs. However, such hubs were extremely rare (I have not seen any in my life), you could not daisy chain them at all (you had to build your whole network using a single 1Gbps hub and that was it), and understandably, with all their disadvantages when compared to switches, I doubt that any networking vendor seriously considered manufacturing 1Gbps Ethernet hubs. Even with half duplex, a 1Gbps Ethernet would still use all 4 pairs in a TP cable.

I case of Ethernet the speed negotiation happens by the exchange of  FLP's which are electrical signals. What is it in case of fibre media ? 

To my best knowledge, fiber transceivers are always manufactured only for a single mode of operation (speed, duplex, wavelength, fiber parameters) of operation. For example:

Switch# show int te0/1 capabilities

TenGigabitEthernet0/1

  Model:                 WS-C3560E-24TD

  Type:                  10GBase-LRM

  Speed:                 10000

  Duplex:                full

  Trunk encap. type:     802.1Q,ISL

  Trunk mode:            on,off,desirable,nonegotiate

  Channel:               yes

  Broadcast suppression: percentage(0-100)

  Flowcontrol:           rx-(off,on,desired),tx-(none)

  Fast Start:            yes

  QoS scheduling:        rx-(not configurable on per port basis),

                         tx-(4q3t) (3t: Two configurable values and one fixed.)

  CoS rewrite:           yes

  ToS rewrite:           yes

  UDLD:                  yes

  Inline power:          no

  SPAN:                  source/destination

  PortSecure:            yes

  Dot1x:                 yes

Note there is no alternative listed in the highlighted lines.

Therefore, negotiation is not really an issue here. However, at least in 100Mbps and faster versions, these transceivers exchange continuous streams of symbols (either frames or IDLE symbols) to keep synchronized and to know there is a live device connected.

Best regards,

Peter

[Umesh Shetty]

Thnx Peter,

This would not be a correct assessment of the situation. FastEthernet always uses 2 pairs, one for Rx, the other for Tx. The duplex setting has no influence on what pairs are used. On a single TP cable, there is in fact no way of creating an electrical collision. However, if there was a hub, two stations operating in full duplex would cause the collision in the hub's circuitry. That is why even with half-duplex and 2 distinct pairs for Rx/Tx, Ethernet has to perform CSMA/CD to avoid collisions.

Does that mean that if I connect two routers back to back with a cross over cable both will still use 2 pairs and have no chance of collission and performance degradation?

Also on a Gig port if a connect a 1 Gig fibre port it will be capable of only single speed and sigle duplex and on the same port if I connect a 1 Gig copper SFP it will be capable of operating on 10/100/100 and full or half duplex ?

Thanks in Advance

Umesh

[Peter Paluch · Accepted solution]

Hi Umesh,

Does that mean that if I connect two routers back to back with a cross  over cable both will still use 2 pairs and have no chance of collission  and performance degradation?

Yes, that is correct.

Also on a Gig port if a connect a 1 Gig fibre port it will be capable of  only single speed and sigle duplex and on the same port if I connect a 1  Gig copper SFP it will be capable of operating on 10/100/100 and full  or half duplex ?

Yes, that would be my understanding. The transceivers (SFP modules) are dealing with physical layer issues, and the speed and duplex are physical layer properties. Depending on the transceiver, various speeds, duplex modes and their combinations may or may not be allowed.

Best regards,

Peter

[Umesh Shetty]

Thnx Peter,

I think I only asked half part of my question, what I meant to ask was two routers connected back to back with a cross over cable and running on half duplex. Will there be no collission? and becasue both the nic's  are operating at half duplex will they perform CSMA/CD ?

There won't be any problem either if one end is half and the other is full , in that case only one end will perform CSMA/CD.

So the only difference between  two systems connected back to back on half duplex at both ends and two systems as operating at full duplex at both end would be only that the systems operating at half duplex would perform CSMA/CD and hence would not offer the performance that full duplex would offer.

Problem only occurs when multiple systems are connected on a hub and all are set to full duplex and all sending at the same time. The hub when it receives transit packets from multiple ports in and out will sense collission and as a result drops some packets.

Can you share your thoughts please.

Thanks in Advance

Regards

Umesh

[Peter Paluch · Accepted solution]

Hi Umesh,

what I meant to ask was two routers connected back to back with a cross  over cable and running on half duplex. Will there be no collission? and  becasue both the nic's  are operating at half duplex will they perform  CSMA/CD ?

Aaah, I see. Two routers connected back to back with a crossover cable and operating in half duplex may cause collisions. Please keep in mind that on TP with Ethernet and FastEthernet (10 and 100 Mbps), there is no possibility of creating an electrical collision on the cable itself, as the Tx and Rx pathways are distinct. A collision here is understood simply as an event of a network card suddenly receiving data when it is transmitting data itself. As the half duplex mandates that the network card either sends data or receives data but not both at the same time, a simple occurence of data on the Rx pair while the card is sending its own data on the Tx pair essentially means that there are two cards sending data at the same time, and this is considered a collision. Even with two routers interconnected by a crossover cable, it may happen that both routers need to send data at the same time, so they both start listening to the media, sense no carrier, and start sending, creating the collision as explained above.

So - in the scenario you just described, there can be collisions, and the NICs will be doing the CSMA/CD.

There won't be any problem either if one end is half and the other is full , in that case only one end will perform CSMA/CD.

On the contrary, there will be lots of problems. The full duplex NIC will not perform CSMA/CD and will simply start sending data as soon as they are ready, regardless of whether the NIC is receiving data from the other side or not. The half duplex NIC, on the contrary, expects that when it is sending data, no one else is. The duplex mismatch is deadly for the half duplex NIC - it will experience numerous collisions because the full duplex NIC simply does not care if the half duplex NIC is sending data or not, and it will keep interrupting it whenever it has some data to send.

So the only difference between  two systems connected back to back on  half duplex at both ends and two systems as operating at full duplex at  both end would be only that the systems operating at half duplex would  perform CSMA/CD and hence would not offer the performance that full  duplex would offer.

No, not true. Two systems connected back to back on half duplex will revert to CSMA/CD and they will be prone to collisions. Only having them both operate in full duplex will result in a collision-free operation.

Problem only occurs when multiple systems are connected on a hub and all  are set to full duplex and all sending at the same time. The hub when  it receives transit packets from multiple ports in and out will sense  collission and as a result drops some packets.

Well, think of it this way: consider three stations on a hub, A, B and C, all thinking they can do full duplex. Now when A and B start sending data to each other, the hub would need to copy the transmission to C - but as there are two transmissions taking place at the same time, C would be getting something like A+B, which is a true collision. C would not understand a thing, and the hub would know it is experiencing collision.

Best regards,

Peter

[Umesh Shetty]

Thnx Peter,

That was really helpful indeed !!!

Another question, I recently had a situation where we connected a new Dell server on my 3750X switch using a Cat 6 patch cable. Both the switchport and the server were set to aut negotiate but still the server end if the connection would negotiate to Speed 100 instead of speed 1000. Hardcoding worked well, but we wanted it to work on auto. We then simply changed the patch cable and it worked well both stations auto negotiated to 1000.  I know that speed aoto negotiation works through the exchange of electrical signals. What do you think can go wrong with the physical cable causing this not to work.Is it just the cable going bad and causing degradation in the actual signals.

I was also interested in knowing that when two routers with Gig ports are connected back to back and one is hardcoded to speed 100 and the other is left to auto negotiate, it will negotiate to speed 100 through reception of electrical signals.

I wanted to know how the electrical signals actually change when one end is set to 100 and the very next moment when the speed is changed to 1000 the far end auto negotiates to 1000. What in the signals ? and how do the they change when one end changes ?

I was also going through the auto mdix feature and one of the docs at the below mentioned link says "Automatic medium-dependent interface crossover (auto-MDIX) is enabled by default. When you enable auto-MDIX, you must also set the interface speed and duplex to auto so that the feature operates correctly."   Could you help explain why speed/duplex auto is required for auto mdix to work correctly ?

Also when connecting two switches using straight cable and auto mdix enabled one end of the switchport will auto sense the cable type and swap the Tx / Rx pair ie start using 1and 2 for Tx and 3 and 6 for Rx , is this right ?

Thanks in Advance

Umesh 

[Peter Paluch · Accepted solution]

Hi Umesh,

I recently had a situation where we connected a new Dell server on my  3750X switch using a Cat 6 patch cable. Both the switchport and the  server were set to aut negotiate but still the server end if the  connection would negotiate to Speed 100 instead of speed 1000.  Hardcoding worked well, but we wanted it to work on auto. We then simply  changed the patch cable and it worked well both stations auto  negotiated to 1000.

If hardcoding 1Gbps worked fine while autonegotiation resulted in only 100Mbps operation then I am confused. It is a known fact that for 1Gbps operation, the cable either has to be a properly pinned straight-through cable, or a cross-over cable that swaps all four pairs (not just the green and orange as seen in 10/100Mbps cross-over cables). If the cable does not conform to either of these requirements, the autonegotiation will end up with 100Mbps and hardcoding the 1Gbps results in a totally inoperable link. This happens quite often with cross-over cables that keep the blue and brown pairs in place. Such cables are fine with 10/100Mbps Ethernet but are unsuitable with Gigabit Ethernet. However, I do not quite understand how the cable could have worked with hardcoded speed when the autonegotiation was able to negotiate only 100Mbps.

Can you confirm if the cable was indeed a cross-over cable, and if so, was it a fully crossed cable?

I wanted to know how the electrical signals actually change when one end  is set to 100 and the very next moment when the speed is changed to  1000 the far end auto negotiates to 1000. What in the signals ? and how  do the they change when one end changes ?

The autonegotiation itself is not doing any cable testing. It just uses different pairs to advertise supported speed and duplex settings of both devices as a series of double-byte words (called pages), and they choose the best mutually supported variant. If the cable is subsequently unable to support such speed (because, for example, the pairing has not been followed according to EIA/TIA-568), the link will start flapping. So once again, the autonegotiation is not just about sensing some signals - it is actually about exchanging double-byte words that encode supported capabilities of its sender. You may be interested in reading the following documents:

http://www.n-tron.com/pdf/ethernet_autonegotationwp.pdf

http://www.ethermanage.com/ethernet/pdf/dell-auto-neg.pdf

https://www.iol.unh.edu/services/testing/ethernet/knowledgebase/Clause_28_Auto-Negotiation.pdf

They do cover the autonegotiation in very nice and readable terms.

I was also going through the auto mdix feature and one of the docs at the below mentioned link says "Automatic  medium-dependent interface crossover (auto-MDIX) is enabled by default.  When you enable auto-MDIX, you must also set the interface speed and  duplex to auto so that the feature operates correctly."   Could you help  explain why speed/duplex auto is required for auto mdix to work  correctly ?

I wish I knew There are some documents I will need to study further. So far, I do not see any dependence between the autonegotiation and the auto-MDIX but I can be mistaken here. Give me a couple of days, I will try to find out.

Also when connecting two switches using  straight cable and auto mdix enabled one end of the switchport will auto  sense the cable type and swap the Tx / Rx pair ie start using 1and 2  for Tx and 3 and 6 for Rx , is this right ?

Yes, that is correct.

Best regards,

Peter

[Umesh Shetty]

Thnx Peter for your inputs,

I finally tried testing the 4 pair crossover thing. I used a crossover cable with only the orange and green pairs crossed and the blue and green in their original positions. I connected two 3750G-24TS switches back to back using that cable with auto mdix disabled and speed and duplex set to auto negotiate at both ends. So theoretically becasue only 2 pairs are crossed on the cable and auto mdix is disabled the ports at both ends must negotiate to 100 ie the next best possible speed on 2 pairs. But strangely the ports negotiated to 1000 Mbps :O . Need your help to understand if I am missing something here

Thanks in Advance

Umesh

[Joseph W. Doherty · Accepted solution]

Disclaimer

The   Author of this posting offers the information contained within this   posting without consideration and with the reader's understanding that   there's no implied or expressed suitability or fitness for any purpose.   Information provided is for informational purposes only and should not   be construed as rendering professional advice of any kind. Usage of  this  posting's information is solely at reader's own risk.

Liability Disclaimer

In   no event shall Author be liable for any damages whatsoever (including,   without limitation, damages for loss of use, data or profit) arising  out  of the use or inability to use the posting's information even if  Author  has been advised of the possibility of such damage.

Posting

To add (hopefully) to the information already provided by Peter . . .

Fastethernet at full duplex uses 2 pairs or the wire RX (wire 1 and 2) and TX (wires 3 and 6) . Incase the Fastethernet port is set/negotiated to half duplex will it continue to use only 1 pair to both send and receive and hence needing to do CSMA/CD to avoid collissions.

10BaseT transmits and receives, on its TX/RX, whether running half duplex or full duplex.  The "trick" is, on hubs host TX is sent to every host's RX, including the sender's while on switches, the frame is captured, buffered and then just sent to selective ports' host RX, always excluding the sender. On the host, half duplex vs. full duplex change the host's expectations such that is shouldn't see any collision on its RX and it "knows" it can transmit on TX even while "seeing" traffic on its RX.  (Basically, [half duplex] BaseT initially mimics Base2 and Base5 media sharing, but BaseT always was different media for TX and RX.  Full duplex, takes advantage of this.)

Many Gig etherne ports are capable of 10/100/100 interms of speed and half and full interms of duplex. So is the half duplex possible only when the speed is 10 or 100. Or with a speed of 1000 also half duplex can be acheived ? If yes how many pairs of the wire will it use ?

I think I recall BaseT Gig Ethernet requires full-duplex.  If the device allowed gig half-duplex, two pairs would still be required to support gig bandwidth.  It's how TX/RX are managed (as I tried to explain, above) that differs.

I case of Ethernet the speed negotiation happens by the exchange of FLP's which are electrical signals. What is it in case of fibre media ?

As noted by Peter, fiber is generally single speed (although ports like SFP+ support different speed optics).

Technically, I'm sure you could make multispeed optics, but "needs" for copper and fiber multispeed differ.

Copper multispeed was developed to make migration much easier between Ethernet and FastEthernet (and later GigEthernet) although I believe I recall some early device FastEthernet didn't support Ethernet on the port.  Also, with FastEthernet full duplex was optional, and I don't think it was actually really defined to be used for original Ethernet usage (although it was retrofitted, by hardware vendors, to "later" Ethernet).  ("Auto" didn't always work well during the early years too.)

Fiber usage (i.e. port counts) was much, much less, and with fiber other consideration apply such as kind of fiber being used (although copper has some similar issues with different "categories" of cable, e.g. CAT3 vs. CAT7).  So, suspect hardware vendors never saw much of a demand for this.  Interestingly, with DWDM optics, some are "tunable", most not.

[Peter Paluch]

Hi Joe,

Thanks for joining! Something of what you wrote caught my attention.

10BaseT transmits and receives, on its TX/RX, whether running half  duplex or full duplex.  The "trick" is, on hubs host TX is sent to every  host's RX, including the sender's

Are you absolutely sure by this? Would this not be perceived as a collision by the sender?

Think of this scenario: Two computers connected back to back with a crossover cable, both configured for half duplex, and causing a collision because both try to transmit at the same time. How is this different from a situation with a computer and a hub running in half duplex, with the hub sending the Tx data out through all Rx pairs including the original sender, thereby again appearing as a colliding device?

It was always my understanding that even with a hub, a received signal is propagated out the Rx pairs over all remaining connected ports except the port where the signal is received. Echoing it back to the sender would not be distinguishable from an ordinary collision caused by two devices talking in the same time.

Best regards,

Peter

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Are you absolutely sure by this? Would this not be perceived as a collision by the sender?

Peter, no I'm not 100% certain for 10BaseT.  As far as the sender is concerned, it could "monitor" what's being received while transmitting.  The latter would allow the sender to detect a collision as it would "see" its transmission has been corrupted.  (As 10Base2 and 10Base5 use a shared medium, an Ethernet sender must be able to deal with this, agree?  [For shared media Ethernet, collisions are often described as senders detecting abnormal voltages - from frames from more than one sender overlapping.])

Specific to 10BaseT, on the ingress port, I just tried to research this, and couldn't find clarity.  I did see mention that certain hubs can conform to Ethernet with slightly different behaviors, such as some (the hub themselves) generating jam signals when they detect a collision.

Think of this scenario: Two computers connected back to back with a crossover cable, both configured for half duplex, and causing a collision because both try to transmit at the same time.

Yes, that's an excellent scenario.  How do the two back-to-back hosts know there's a collision when in fact they are physically full duplex?  It could be when they receive any signal while transmitting or it could be what kind of signal they receive while transmitting.  Again, I couldn't quickly ascertain which for 10BaseT.