2
\$\begingroup\$

I did some searching on how propagation delay in ethernet cables is affected by the cable length. The answer seems to be that the delay is negligible until the cable length reaches a kind of threshold value which imposes a limit on the maximum length of ethernet before it becomes too unreliable. However, propagation delay is distance/speed, so technically a longer cable should have a longer delay. I am assuming the cables have the same diameter.

My guess is that since information is continuously sent in a stream, while it takes longer to initiate the transfer, the rate of data transfer is unaffected. However, the latency would still be slower, which would be a problem. Why is the cable length rarely discussed?

\$\endgroup\$

3 Answers 3

6
\$\begingroup\$

No, there is no threshold as you put it; any length of cable produces a delay and that delay is proportional to cable length. Different cables propagate slower of course and this is largely down to the dielectric of the material between the two conductors (or centre and screen in coax). The higher the capacitance, the slower the speed of propagation.

Regarding maximum length that can be used for a certain data rate yes, there is a "kind of" threshold - basically data gets misshaped the further it has to travel down a cable due to cable losses (resistive and dielectric). At some cable-length and at some data-rate the cable can be deemed to be at the "point of no return" in that statistically the number of data errors incurred are too many to warrant further error correction/detection. Time to get a better cable or different modulation scheme!

Cable length is quite commonly discussed where I work for the reasons highlighted above.

\$\endgroup\$
3
  • \$\begingroup\$ @apalopohapa my answer was only concerned with the physics of the cable rather than the protocol requirements \$\endgroup\$
    – Andy aka
    Jun 4, 2014 at 17:04
  • \$\begingroup\$ I just noticed this after writing a protocol based answer. :( \$\endgroup\$
    – Hennes
    Jun 4, 2014 at 17:13
  • \$\begingroup\$ Rule of thumb: 2/3 speed of light, or roughly 5ns per meter of cable. A short frame at 100Mbit/s takes about 6us to transmit or receive. Yet, there is a protocol called PTP which synchronizes clocks over Ethernet to sub-microsecond precision where the cable delay is actually measured. \$\endgroup\$
    – maxy
    Jun 8, 2014 at 15:29
9
\$\begingroup\$

As Andy aka already wrote, any length of cable produces a delay. However by limiting the length you know what the maximum delay is and you can correct for that. Ethernet does this by specifying a minimum frame size in combination with a maximum cable length.

Thus ensure that even two computers on the ends of the cable (and thus the furthest away from each other as allowed) will see that the other is transmitting before they finish their own frame.

Imagine the following setup with two computers and ancient coax: enter image description here

Both computer #1 and #3 want to transmit something.
Both check the cable and see noone else transmitting.
Both start to transmit their own small frame which will travel down the cable at limited speed.

enter image description here

Neither computer 1 or computer 3 will see the others signal while they are transmitting.
Both will receive a readable frame from the other.

however computer 2 will see an unidentifiable mess rather than one or more valid frames.

The solution employed by Ethernet is to limit cable length to about 200 meters and make a framesize of at least 64 char mandatory. (If the frame is shorter it will be padded).

That way any transmitting from one edge of the network will reach the other side of the network. Any computer (well, any NIC) wanting to transmit needs to check if the line is clear and stays clear from the start of transmission until it has send at least the min. frame size.


Finally getting to your question on Why is the cable length rarely discussed? (in combination with the tag Ethernet):

From ancient times the min. frame size and max. cable length have been defined for Ethernet. This is a standard. Even with UTP we still adhere to those standards. This makes the values always the same and not worth discussing.

\$\endgroup\$
5
  • \$\begingroup\$ Thanks for this; as Andy mentioned I was looking for a purely physics based answer rather than protocol based so I've accepted his answer \$\endgroup\$
    – ddriver1
    Jun 4, 2014 at 22:28
  • 3
    \$\begingroup\$ The description here is for half-duplex Ethernet. While it is accurate, it's not how Ethernet typically works today. Today's Ethernet cables connect only between two devices (e.g. PC and Switch) full duplex, there is no collision when both devices transmit at the same time. Switch forwaring creates much more delay than cable signal propagation. \$\endgroup\$
    – maxy
    Jun 8, 2014 at 15:11
  • \$\begingroup\$ @Hennes, What software do you use to draw that? \$\endgroup\$
    – Pacerier
    Mar 11, 2017 at 0:07
  • \$\begingroup\$ @maxy, What about big lans? \$\endgroup\$
    – Pacerier
    Mar 11, 2017 at 0:07
  • \$\begingroup\$ That was plain old ancient mspaint.exe. :) As for big LAN, nothing essentially changes. If you add more NICs/computers then you might need to wait more often until the line is free and you have bigger risks on collisions. Or with modern switches: You and the switch are the only ones on your particular cable segment. And in that case there is either a very low risk of collision or you even more the protocol to FD, basically changing the pair from 'line is busy' to a second data path. So both devices get their own 100Mbit (or whatever) lines. \$\endgroup\$
    – Hennes
    Mar 11, 2017 at 12:16
1
\$\begingroup\$

Any length of cable imposes a delay which is roughly proportional to it's length. There will also be a degradation in signal integrity, again getting worse with length.

If the signal integrity degrades too much then the receiver will not be able to correctly interpret the signals. For coax and twisted pair Ethernet this is what limits the length of a single "segment".

Large propagation delays can cause incorrect collision detection in a CSMA/CD system. Specifically there are problems if the propagation delays grows larger than the minimum packet length. With a single segment of coax or twisted pair Ethernet that meets the length requirements for signal integrity you won't run into this problem. However you can run into it in two cases.

  1. chains of repeaters/hubs (NOT switches): With 10 megabit ethernet the rule was you should have no more than 4 repeaters between the most distant segments in a collision domain. With 100 megabit Ethernet this reduced to 2[1].
  2. fibre optic links: The fibre physical layers can maintain signal integrity over a much longer distance. Therefore long fibre links cannot be correctly operated in half duplex mode.

Full duplex links do not use CSMA/CD. Switch ports running in half duplex mode do use CSMA/CD but the collision domain does not cross the switch.

So basically as long as you make sure the length limits for the physical layer are respected, make sure you use switches rather than hubs and make sure any long distance fibre links are running in full duplex mode then you are fine.

[1] With gigabit the rules stayed the same as with 100 megabit because they increased the minimum packet size for half duplex gigabit links but it was irrelevant in practice because gigabit hubs are like hens teeth.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.