Copper or Glass for Low Latency?

Question

I'd like to preface the main content of this post with the following: I understand this question may be equally suited to the Physics Stack Exchange, and so I have posted on there as well, but maybe a network engineer or computer science wizard may also be able to provide some illuminating knowledge.

Which medium allows for lower latency in data transfer, copper or fiber?

It is a tie (sort of). The transfer of information through electromagnetic waves, or the signal, has a velocity of approximately 200,000 km/s in both copper and glass.

A couple key advantages to optical transmission are it is immune to electromagnetic interference, and it has a much larger bandwidth, over 1000 times more than electrical transmission.

However, to achieve the lowest possible latency in a network, fiber should be used from the provider's equipment to the customer's network. Copper should be used from the customer's gateway to devices requiring a low-latency connection.

If the latency is the same in both mediums, then why not use just one medium, say copper?

The reason for this is that the distance from the customer's network to the telco's equipment is usually large enough that if the link was copper, it would require a signal repeater at some point to ensure the signal doesn't become too weak for the telco's equipment to register. Because of the additional device, the signal repeater, that data has to go through, it will take longer for the data to reach its destination than if it had traveled through just fiber. Optical transmission of data through fiber allows for much farther transmission of data due to the signal remaining stronger for longer without the need for a signal repeater to be used anywhere close to the amount needed in long distance electrical transmission (I think one would be needed for approximately every 300' in copper, but single-mode fiber can carry a signal for miles before needing to be repeated).

The reason why copper should be used in LANs is that many devices cannot yet work directly with light. Upon an optical signal entering a device, it has to be converted into an electrical signal. The opposite is true for a signal exiting a device. These conversions add time to the process. The points of conversion are essentially additional devices where the signal must pass through and be processed.

Answer 1

You already provide most answers.

Propagation delay: apart from (good) coax, copper and fiber are nearly equal. Decent twisted pair is slightly faster than fiber, high-quality TP even more so - but nothing really relevant.

If bandwidth was a non-issue, propagation speed on 10BASE5 is faster than fiber (a whooping 320 ns ahead over 500 m!) - but the ancient hardware surely is a show-stopper.

Conversion delay: (simple) line codes need to be converted in any case, so there's no real difference when assuming integrated converters or modules (SFP). External converters cost extra, of course.

However, copper may require more elaborate line codes, increasing latency. For instance, 10GBASE-T requires some additional 0.5-1.5 μs (micro seconds) in comparison to 10GBASE-R or DAC (also running -R PCS code).

Long-haul, voice-grade copper requires very elaborate encoding, depending on the distance and speed. VDSL on short links may start at 3-5 ms (milli seconds) but interleaved ADSL on long loops may exceed 60 ms.

Fiber in comparison runs for 100 km or so with gigabit speed or faster, with simple line code, on passive fiber. Longer links require boosting, but with optical amplifiers (EDFAs) that doesn't increase latency.

In a nutshell: Decent twisted pair is good for short distance (max. 100 m) and up to 10 Gbit/s, anything faster or longer requires fiber. DAC is high speed but very short distance. Low-delay requirements demand fiber or DAC.

Answer 2

Latency comes from several places including.

1. Propagation delay: How long does it take for a signal at one end of a cable to reach the other. Fiber and twisted pair are pretty similar. Some coax is lower. Free space links (microwave or optical) are lower still.
2. Processing delay: Having received a packet how long does it take the equipment to decide what to do with it.
3. Queing delay. How long does the packet have to wait for a transmission slot.
4. Serialisation delay: How long does it take to go from a packet in a buffer to a packet in a line-rate data stream. Most devices operate in a "store and forward" manner so this price is be paid at each hop. Sometimes "cut through" forwarding is possible but that has problems of it's own and can't be used in all circumstances.
5. Coding delay. How much latency does the line coding scheme introduce. Copper often requires more elaborate coding schemes which add more delay.

While data rate does not directly effect propagation delay, it very much does affect queuing delay, serialisation delay and coding delay. If your network runs at 10 megabits per second then a 1500 byte packet takes 1.2 milliseconds to serialise. If your network runs at 10 gigabits per second it only take 1.2 microseconds to serialise.

---

The reason why copper should be used in LANs is that many devices cannot yet work directly with light. Upon an optical signal entering a device, it has to be converted into an electrical signal. The opposite is true for a signal exiting a device. These conversions add time to the process.

As speeds increase, the idea of what is a "long distance" shrinks. 10 gigabit fiber offers lower latency than 10GBASE-T. SFP+ direct attach is comparable to fiber in latency but is limited to very short distances.

At speeds beyond 10G, BASE-T simply doesn't seem to be an option right now. The IEEE has published standards for 25GBASE-T and 40GBASE-T but I haven't see any evidence of either of them actually showing up in products.

That said, latency on the LAN is usually negligible compared to latency on the WAN.