Finally, I don't know really what is collision. Could we exchange some words about this?
Let's suppose 2 stations connected through a 100Base-TX link.
- If the 2 NIC are configured in full duplex, the collision detection is inactive.
- If the 2 NIC are configured in half duplex, the collision detection is active although there is no possible collision on the medium (the 2 twisted pair). At which other Ethernet level the collision can occur if the 2 stations send a frame at the same time?
Not correct. A collision is a condition, on a half-duplex Ethernet, where two or more stations are attempting to transmit a frame at the same time. In the case of a two station, point-to-point half-duplex link, if both stations have a frame queued for transmission at the same time, they will both attempt to transmit it, a collision will occur, the stations will detect the collision, and then backoff and retry.
Clearly, in the simple point-to-point configuration, the stations
*could* have been configured for full-duplex operation, in which case the event of "I am transmitting and receiving at the same time" would not constitute a collision, and no backoff-and-retry would be necessary.
Collisions occur and are detected in the Physical Layer; the reaction (backoff and retry) occurs at the MAC sublayer.
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
For coaxial ethernet it might be more obvious, as there is only one cable. For twisted pair, collision detection is much easier. When a frame is seen on the receive port during transmission, a collision is detected and processed. With UTP, a collision domain may include other hosts connected through repeaters. Even with only a two host system, some NICs could share logic between the receiver and transmitter, and be unable to operate in full duplex mode.
There is another case that you didn't mention:
One NIC configured full duplex connected to one configured as half duplex. This can result in very slow network operation as one tries to retransmit data, and the other doesn't realize that its data is being lost.
It seems difficult to use the word frame here... except if the opposite station is in full duplex. I would say "when a signal...", and perhaps not any signal because we have idles between frames, in
100Base-TX.
So, this formulation would seem more complex to write than those chosen by Rich. Could you try to write how the condition of collision is detected by the NIC, in terms of voltage, power...
Could you detail a little bit more? Is it at the PMD / PMA / PCS / Reconciliation or upper part.
Yes; in the situation of two half-duplex NICs connected via
10BASE-T/100BASE-TX, this is exactly the case. The received signal carries a valid frame, but the NIC declares a collision anyway because in half-duplex, transmitting while receiving ist verboten.
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
The electrical details of how collisions are detected depends on the specific variant of Ethernet being used. In 10BASE-T, it would be the reception of signals exceeding a specified threshold for a specified period of time (corresponding to "normal" Ethernet encodings); in
100BASE-TX, it is (roughly) the reception of non-IDLE signals for a specified period of time, etc.
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
I had through of a few words to use there, and none seemed exactly right. Frames do exist in half duplex, though the beginning of the preamble is likely to be the actual signal that does it.
For coaxial ethernet it is necessary to compute the voltage based on cable resistance, terminators, and the current supplied by the transceiver worst case for one station transmitting and for two.
For UTP it is much simpler, with separate wires for send and receive.
In years past the logic that could fit on an IC was much less, and one might have tried to conserve. The one I was thinking of at the time wa the CRC shift register, but there may be more.
Yes. There is one interesting case. After waiting the appropriate inter frame gap and backoff delay (if after a previous collision) and one is ready to transmit, one does not defer if another station is seen to be transmitting. (That is, within the same backoff slot window, if any.) Doing this removes the advantage one might have with a slightly faster, but still within tolerance, clock.
Otherwise, some of the complication is necessary so that repeaters work properly, and all stations detect a collision when one actually occurred.
Ich denke, da Sie danken Ihnen bedeuteten. (I think you meant "Thank you", not "please".)
Correct.
Well, it's *possible*, but a given device may not support half-duplex mode. It is perfectly permissible under the standard to build a device that is incapable of operating in half-duplex mode, even if the medium supports it; this is quite common for Gigabit Ethernet.)
I'm not sure what you mean by "half duplex violation".
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
Thanks for that good comment. We can perhaps trace back to the beginning.
- Before 1990, 10 Mbit/s Ethernet networks have been only half-duplex.
- From 1990 to 1998 the standard offered, with 10 Mbit/s and 100 Mbit/ s, half-duplex only or the both modes. Probably users were stimulated for transfering (migrating) there networks from half-duplex to full- duplex.
- From 1998 to 2002, with 1G, the standard offered the both modes. And perhaps some users could have still held, temporarily, some old half- duplex 10/100 interfaces mixed to 1G.
- Since 2002, with 10G, the full duplex operation is only defined.
Sorry, I meant half-duplex infringement, transgression, infraction (...by non idle signals received while transmiting).
Under the *standard*, Ethernet networks have been half-duplex-only until
1997, when IEEE 802.3X was approved. This was true both for 10 Mb/s and
100 Mb/s operation. Prior to 1997, some vendors offered proprietary, non-standard, full-duplex versions of Ethernet, both at 10 Mb/s and 100 Mb/s.
In 1997, the standards body approved 802.3X, which provided for full-duplex operation of Ethernet at all data rates. Vendors were (and still are) free to build devices that could operate in half-duplex-only, full-duplex-only, or both modes.
In practice, there are no devices operating at 1000 Mb/s in half-duplex mode; I know of no commercially-available repeaters for Gigabit Ethernet. (And yes, I am aware of the *very limited* use of half-duplex Gigabit Ethernet for certain Cisco inter-chassis point-to-point links.)
Operation at 10 Gb/s requires full-duplex mode only.
I still don't get it. Collisions are not a "violation" of anything; they are the ordinary method by which channel arbitration is accomplished in half-duplex mode. In full-duplex mode, no channel arbitration is required (and hence, no collisions are noted).
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
(snip) Do they have to state which one they are doing? Might one be surprised to find a 100baseTX device that wouldn't work on a half duplex system?
Do you know of any gigabit NICs that implement half duplex, just in case they were ever connected to a repeater? Presumably many implement it when in 10baseT or 100baseTX mode so much of the logic would already be there.
The standards don't require that vendors "state" anything, however any manufacturer will surely have a data sheet that says what the product does, be it half/full/both. In practice, I don't know of any 10/100 Mb/s devices that cannot operate in either mode; it's full-duplex-only Gigabit that is the important case.
I'm not up on specific product offerings these days, sorry.
-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX
Thanks very much for your time. We appreciate your words.
And you were very close to this standard body, because you were the task force chair and the task force editor of the 802.3X !... so we have to read twice and more your lines.
I had a similar remark to Glen by reading you, Rich. With Gigabit Ethernet devices only in full-duplex mode, it's impossible to buy a Gigabit Ethernet repeater (a pure hub)... and nobody needs such a device.
I realize also we had, in a repeater, all the Ethernet interfaces from the same rate (either all 10 Mbit/s ports, or all 100 Mbit/s ports). But, perhaps, the possibily to have different types of interfaces in the same rate (i.e a device with 3 ports in 100Base-TX and 1 port in
100Base-FX). Is-it correct from the point of view of the standard.
Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.