Why Is Ethernet Limited to 100 Metres? Channel, Permanent Link and the 90 Plus 10 Rule

Every cabling technician knows the number. One hundred metres. It gets quoted like a law of physics, and it is almost never questioned.

It is worth questioning, because the number is not a property of Ethernet at all. It is a property of the cabling standard, and underneath it is not a distance rule. It is a loss budget. Once you see it that way, a lot of things that look arbitrary start making sense: why 90 and not 100, why your long patch cords cost you more than their own length, why a run that certified in June fails in February, and why 10 Gigabit Ethernet stops dead at 37 metres over Category 6 cable that is rated to 100.

The Short Answer

Three numbers, and they are not interchangeable.

ISO/IEC 11801 puts it plainly. The physical length of the channel shall not exceed 100 m, and the physical length of the fixed horizontal cable shall not exceed 90 m. When the total length of patch, equipment and work area cords exceeds 10 m, the allowed length of the fixed horizontal cable has to come down.

That last sentence is the one people miss. The 90 and the 10 are not two separate allowances that you can spend independently. They trade against each other, and they do not trade one for one.

Channel Versus Permanent Link

These two words get used loosely on site, and the difference matters more than almost anything else in this article.

The permanent link is the fixed cable plus the connector at each end. Two mated connections, 90 metres maximum. It is what gets installed during construction, and it is what an installer certifies with a tester before the ceiling closes. It is called permanent because nobody is going to change it without pulling the building apart.

The channel is the whole path the signal takes. It adds the equipment cord at the switch end and the work area cord at the device end, and it can carry up to four mated connections. One hundred metres maximum. This is the thing your electronics actually has to work across, and it is the thing nobody tests, because at the time of testing the cords do not exist yet.

That gap is where most real world failures live. A permanent link certificate proves the wall cable is good. It says nothing about the 15 metre cord somebody later ran under a desk because the outlet was in an inconvenient spot.

ISO/IEC 11801 adds a few configuration limits that are worth committing to memory:

• A consolidation point has to sit at least 15 metres from the floor distributor.
• Where a multi user outlet assembly is used, the work area cord should not exceed 20 metres.
• Patch cords and jumper cables should not exceed 5 metres.

Why 100 Metres and Not 150?

Here is the part that is genuinely interesting. The standard does not start with a distance and work out the loss. It starts with the loss and works out the distance.

A receiver can only tolerate so much attenuation before it can no longer recover the signal. The cabling standard fixes that figure, publishes it as a maximum channel insertion loss, and then the cable and connector specifications are set so that 100 metres of qualifying components fits inside it. The distance is the output of the calculation, not the input.

These are the published channel insertion loss limits at each class top frequency:

Note that the loss limit gets larger for the better cable, not smaller. That surprises people. It is because the better cable is being asked to work at a much higher frequency, and copper attenuation climbs roughly with the square root of frequency. Category 6 at 250 MHz is a far harder job than Category 5e at 100 MHz, so it is allowed a bigger loss number while still delivering more.

The channel loss model that sits behind those figures is worth seeing, because it explains the 90 plus 10 split directly:

channel insertion loss = 1.05 × (loss of 100 m of cable) + 4 × (loss of one connector) + a device allowance

The 1.05 is the whole story. Flexible stranded cord has noticeably higher loss per metre than the solid conductor cable in the wall, so the model inflates the cable term by five percent to pay for the cords. And the 4 is the four mated connections a channel is allowed.

This is why a metre of cord is not worth a metre of cable. You are not spending distance. You are spending decibels, and cord spends them faster.

The Cord Trade, and Where 90 Actually Comes From

ANSI/TIA/EIA-568-B gives the trade as an equation:

C = (102 − H) / (1 + D)

where C is the maximum combined length of the work area cable, equipment cable and patch cords, H is the length of the horizontal cable, and D is the de-rating factor for the cord type. D is 0.2 for 24 AWG cord and 0.5 for 26 AWG cord, reflecting that those cords are 20 percent and 50 percent lossier per metre than the fixed cable.

Now put the standard numbers in. With 24 AWG cords, D is 0.2, so ten metres of cord gives H = 102 − 1.2 × 10 = 90.

The famous 90 metres is not a separate rule handed down from anywhere. It is what falls out of the equation when you assume ten metres of ordinary 24 AWG cord. Change the cord assumption and the 90 changes with it.

TIA publishes the same relationship as a table, and it is a useful sanity check on the equation:

What Actually Happens Past 100 Metres

Not what most people expect. There is no trip point at 101 metres.

Modern twisted pair Ethernet uses adaptive equalisation and forward error correction, so a slightly over length channel usually still passes traffic. What you get instead is a link that has quietly spent its margin. The transition looks like this in practice:

1. The link trains and looks completely healthy.
2. Error correction starts working harder, correcting a rising number of symbol errors that never surface as an interface counter.
3. Retransmissions climb, throughput sags, and latency gets erratic under load.
4. On a bad day the link renegotiates to a lower speed, or drops.

That progression is why over length runs are so often blamed on the switch, the patch lead or the device. The cabling passes a ping test perfectly. It fails a busy Monday morning.

It is also why the honest answer to “can I run 120 metres?” is not “no, it will not work”. It is “it will probably work today, on your bench, at 22 degrees, with no traffic on it, and you will have no margin left for anything else that goes wrong”. Nothing in the standard is holding you back from trying. Everything in the standard is telling you what you are giving up.

The Category Question: How Far Does Each One Really Go?

This is where the single number breaks down completely, because reach depends on the application, not just the cable.

The Category 6 case deserves the detail. TIA TSB-155 states that 10GBASE-T should operate over channel lengths of up to 37 metres of Category 6 cabling, and should operate between 37 and 55 metres depending on the alien crosstalk environment. Past 55 metres, mitigation is required.

The reason is not attenuation. It is alien crosstalk, which is the coupling between one cable and its neighbours in the same bundle or tray. A Category 6 cable on its own has plenty of headroom. Fifty of them cable tied together in a riser do not. That is exactly why Category 6A exists, and why Category 6A is specified with alien crosstalk performance that Category 6 never was.

Two practical consequences follow:

• Retrofitting 10 Gigabit onto existing Category 6 is a bundle problem, not a cable problem. How the cable was dressed years ago decides whether it works, and no one wrote that down.
• If you cannot make 10 Gigabit reach, 2.5GBASE-T and 5GBASE-T exist for exactly this situation. IEEE 802.3bz was written to get more out of installed Category 5e and Category 6, both to the full 100 metres.

Category 8 is the other end of the scale and breaks the pattern entirely. It runs to 2000 MHz and is limited to a two connector channel of 30 metres, with 24 metres of horizontal cable and six metres of cords. It also has to be shielded. It is a data centre top of rack product, not a horizontal cabling product, and it is worth being clear about that before someone specifies it for an office floor.

The Australian Problem: Heat

This one gets missed constantly, and in Australia it should not be.

Every length figure quoted so far assumes the cable is at 20 degrees. Copper attenuation rises with temperature, so a hot cable loses more per metre, and a compliant length at 20 degrees is not a compliant length at 50. ISO/IEC 11801 handles this by de-rating the allowed length directly, and the note is explicit. For operating temperatures above 20 degrees, the horizontal length should be reduced by:

• 0.2 percent per degree for screened cables
• 0.4 percent per degree from 20 to 40 degrees, and 0.6 percent per degree above 40 up to 60 degrees, for unscreened cables

Work an example. You have an 85 metre horizontal run in unscreened Category 6A. On the drawing it is compliant, because 85 is less than 90. The cable spends its life in a roof space that reaches 50 degrees on a summer afternoon. The de-rating is 0.4 percent for each of the 20 degrees from 20 to 40, plus 0.6 percent for each of the 10 degrees from 40 to 50, which is 8 plus 6, so 14 percent. The allowed length becomes 90 × 0.86, which is 77.4 metres.

The run was never compliant. It was compliant at 20 degrees, which is a temperature it never sees.

If the cable is also carrying Power over Ethernet, this gets worse, because bundled PoE cables heat themselves. TIA TSB-184-A caps the design temperature rise of the hottest cable in a bundle at 15 degrees above ambient. That rise stacks on top of the ambient temperature before you apply the de-rating above, so a 45 degree ceiling with a fat PoE bundle in it is a 60 degree cable.

The Three Limits Almost Nobody Checks

Length and loss get all the attention. The channel specification contains three more limits that are just as binding, and a field tester will fail a link on any of them.

A couple of observations that put these in perspective.

Propagation delay is closer to binding than it looks. Signal travels along typical twisted pair at a nominal velocity of propagation around 0.65, which is about 0.195 metres per nanosecond. One hundred metres therefore takes roughly 513 nanoseconds against a 555 nanosecond limit. Run the arithmetic backwards and 555 nanoseconds is what 100 metres of cable with a velocity of propagation of about 0.60 would take. The delay limit and the length limit are describing the same channel from two directions.

Loop resistance is where PoE lives. Take a 90 metre permanent link in 24 AWG solid cable at about 0.0842 ohms per metre per conductor. The pair loop is two conductors in series, so roughly 15.2 ohms, and 10 metres of 26 AWG cord adds about another 2.7 ohms, for something near 17.8 ohms at 20 degrees. Copper resistance rises about 0.393 percent per degree, so at 60 degrees the same channel is near 20.6 ohms. Still inside 25, but the margin has gone from healthy to thin, and every volt lost there is a volt the powered device does not get. Note that those figures use solid conductor resistance, and real stranded cords measure higher, so treat the result as a floor rather than an answer.

Delay skew is a cable construction issue you cannot fix later. The four pairs have different twist rates so they do not couple into each other, which means they have different physical lengths inside the same jacket. Cheap cable with badly chosen lay lengths fails skew while passing everything else, and the only remedy is replacing it.

Where This Sits in the Australian Rules

Worth being precise here, because the citation chain is frequently stated wrongly.

AS/CA S009:2020 is the Australian Wiring Rules for customer cabling, and it is the document that carries legal force through the Cabling Provider Rules. It does not contain the insertion loss and length tables. It defers to the generic cabling standards for those.

The 90 metre figure does appear in S009, in Appendix L. It is worth reading exactly what that appendix is, because it is narrower than people assume. Appendix L is titled “Guidance for domestic installations”, it is marked informative, and it says customer cabling in domestic premises should “have a Cable run not exceeding 90 m from the central point to each Telecommunications Outlet”, with a note that reads:

Further length restrictions may apply as a result of heat and specific applications. Refer to AS/NZS 11801.1 and AS 11801.4.

So the note in the Australian standard points at precisely the two effects covered above. For commercial and industrial work, the governing length and performance requirements are in AS/NZS 11801.1, which is the Australian adoption of ISO/IEC 11801-1, with AS 11801.4 covering single tenant homes and AS 11801.3 covering industrial premises.

The short version for anyone writing a specification: cite S009 for the installation and safety rules, and cite the AS/NZS 11801 series for the lengths and the performance classes. Citing S009 as the source of the 90 metre commercial rule is not correct.

Practical Design Rules

• Design the channel, certify the permanent link. They are different numbers. Leave the cord allowance alone rather than spending it during design.
• Budget 90 metres as a ceiling, not a target. If a run is coming out at 88 metres on the drawing, it will not survive the first temperature de-rating or the first long cord.
• Check the ambient temperature of the actual path. Roof spaces, plant rooms and external risers are not 20 degrees, and the de-rating is not optional.
• Add the PoE bundle rise before de-rating, not after. Ambient plus rise is the temperature the cable sees.
• If 10GBASE-T is anywhere in the future, specify Category 6A. The 37 metre Category 6 figure is a real constraint, and it is set by neighbouring cables you cannot control.
• Past about 90 metres of required reach, price the fibre option. Adding a mid span switch to extend copper adds a powered device, a failure point and a maintenance liability. Very often a fibre run is cheaper over the life of the installation.

Frequently Asked Questions

Can Ethernet run further than 100 metres? Over standard twisted pair cabling, not within the standard. The 100 metre channel limit is the point at which compliant performance is guaranteed. Links do frequently pass traffic beyond it, with no margin and no guarantee. If you need more reach, the supported answers are fibre, a mid span switch, or one of the non standard long reach products that trade speed for distance.

Is the 100 metre limit per cable or for the whole run? For the whole run, end to end, cords included. That is what “channel” means. The fixed cable in the wall gets 90 metres of that total.

Does Cat 6A actually go further than Cat 6? For 1, 2.5 and 5 Gigabit, no. Both reach 100 metres. For 10 Gigabit, decisively yes: Category 6A reaches 100 metres where Category 6 is limited to 37 metres, and the difference is alien crosstalk performance rather than loss.

Does PoE reduce the maximum distance? Not directly, since PoE does not change the data reach. It reduces it indirectly through heat, because bundled PoE cables warm themselves and warm cable is de-rated for length. PoE also introduces a separate constraint in the form of voltage drop across the loop resistance, which is a power question rather than a length question.

Why is the permanent link 90 metres and not 95? Because 90 is what the cord equation gives you when you assume 10 metres of ordinary 24 AWG cord. It is a derived figure, not an independent rule.

Do long patch cords really matter that much? Yes, more than their length suggests. Stranded cord is 20 percent lossier per metre than solid cable in 24 AWG, and 50 percent lossier in 26 AWG. Past the 10 metre allowance, each extra metre of cord costs 1.2 or 1.5 metres of wall cable respectively.

Try the Numbers

Working the de-rating, the cord trade and the application reach by hand for every run is exactly the kind of arithmetic that gets skipped on a busy job. The Structured Cabling Channel Planner runs the channel and permanent link checks, the temperature de-rating, the PoE loop resistance and voltage drop, and tells you which applications the resulting channel will actually carry. For a full floor design with racks, pathways and derived run lengths, the Structured Cabling Designer does the same checks across every run in the project.

Notes on Sources

The equations, limits and quoted text above are taken from the standards named below. Insertion loss figures and the horizontal length equations were read from ISO/IEC 11801:2002 and ANSI/TIA/EIA-568-B, and the current editions should be checked before relying on them for a specification. The temperature de-rating example, the loop resistance arithmetic and the propagation delay arithmetic are our own worked calculations from the published coefficients.

• ISO/IEC 11801, Information technology, Generic cabling for customer premises. Table 5 gives the channel insertion loss values at key frequencies, Table 21 gives the horizontal link length equations and the temperature de-rating note, and Clause 7.2.2.2 gives the 100 metre channel and 90 metre horizontal restrictions. Currently published as the ISO/IEC 11801-1:2017 series, adopted in Australia as AS/NZS 11801.1.
• ANSI/TIA/EIA-568-B, Commercial Building Telecommunications Cabling Standard, for the cord length equation C = (102 − H) / (1 + D) and the de-rating factors for 24 AWG and 26 AWG cord.
• TIA TSB-155, Guidelines for the Assessment and Mitigation of Installed Category 6 Cabling to Support 10GBASE-T, for the 37 metre and 55 metre Category 6 reach figures.
• TIA TSB-184-A, for the 15 degree design cap on bundle centre cable temperature rise under PoE.
• IEEE 802.3, Standard for Ethernet, whose clauses define 1000BASE-T, 2.5GBASE-T and 5GBASE-T, 10GBASE-T, and 25GBASE-T and 40GBASE-T.
• AS/CA S009:2020, Installation requirements for customer cabling (Wiring Rules), Appendix L, for the Australian domestic guidance and its pointer to AS/NZS 11801.1 and AS 11801.4.