4
\$\begingroup\$

OCuLink, SATA, USB 3, and Displayport all provide on the order of a few Gbps per signaling pair, and are limited to a cable length of just a couple meters. Ethernet via comparable cabling provides comparable speed, but 50 times the length! For example, 10GBaseT over cat6a can do 100m. What accounts for that extreme difference?

I'm asking in electronics rather than networking since this is a physical link issue, not a networking issue.

EDIT: although latency doesn't appear to be the explanation, people have brought it up, which gave me an idea: why don't all high-speed wired protocols incorporate a latency measurement upon connection (with an upper limit of 200ms for all practical purposes), and add some fixed number of clock cycles to that measurement to establish the retry timeout value for the duration of the connection? Then the timeout is never set higher than necessary, yet all connections ranging from one meter to thousands of kilometers can be accommodated without being limited just by latency. Noise measurement and cancellation, adaptive rate, etc to achieve high speed at long range are more complex, and unnecessary for protocols designed for a few meters or maybe a few tens of meters, but just latency measurement seems very simple and worthwhile, and would avoid arbitrary latency limits like in USB 2 yet without requiring standardizing on an excessively high timeout limit. It just requires a ping, then count the clock cycles until a reply is heard. (To avoid silly misunderstandings: I don't mean ICMP ping; I mean a PHY-to-PHY ping.)

\$\endgroup\$
  • \$\begingroup\$ Compare the price of a 10GBaseT NIC vs. a SATA controller. Even allowing for economy of scale, 10GBaseT is much more expensive to implement than any of the others. \$\endgroup\$ – DoxyLover Oct 28 '14 at 3:59
  • \$\begingroup\$ I believe the most salient difference is that ethernet incorporates highly sophisticated signal processing. \$\endgroup\$ – mkeith Oct 28 '14 at 6:22
  • \$\begingroup\$ On your edit, the main issue is that no-one is being paid to design the perfect protocol. Ideally, everything would use optical transmission - no issues with grounding, galvanic separation, and extremely high theoretical bitrates. However, then you need to design a number of protocols to sit on top of that - again, you'll never get every company and engineer to agree on the best possible protocol. Why not? xkcd.com/927 : that's why... This is more realistic than you might think: every new standard is better but often necessarily not backwards compatible. \$\endgroup\$ – RJR Nov 8 '14 at 23:17
  • \$\begingroup\$ Also on your edit, you are assuming the protocol can deal with the latency. If it can, there is no reason to measure it. You will find a lot of protocols already measure the latency and basically shut off when it is too large. The issue is that larger latencies require significantly different hardware and code to ensure things like clock recovery, correct timing of events, etc. It just isn't worth the money to implement these for systems that generally won't need longer cables. It is easier just to ignore that usecase and constrain the design to lower latencies. \$\endgroup\$ – RJR Nov 8 '14 at 23:28
  • \$\begingroup\$ @RJR - Of course, cost is the explanation for using electrical rather than optical data transmission for short range. The need to modify the encoding, not just the timeout value, does look like a good explanation for not bothering to adjust the timeout to account for latency. I guess I should have asked my edit as a separate question. \$\endgroup\$ – Porthem Nov 8 '14 at 23:54
5
\$\begingroup\$

First, the technocal answer:
For one, USB3 and SATA both use thin stranded twisted pairs (USB3 spec states to mak the cable 'as thin as possible', with 26-34 AWG given as example). 10GbE uses four solid relatively thick twisted pairs (23 AWG minimum). The thicker and solid strands create a larger surface area which in turn means lower resistance to high frequencies (which travel on the our layers of the wire due to skin effect).

Secondly, SATA is designed as a disk interface. It doesn't need to be able to exceed the disk's performance. USB is limited by the fact that USB doesn't support DMA (direct memory access) so for higher speeds, the peripheral's CPU becomes a limiting factor.
Ethernet is designed to use dedicated hardware supporting a multi-point to multi-point topology. It needs to sustain data rates far over that required from a single station. It is therefore also much more expensive.

Lastly, delays become a significant factor at these speeds. Ethernet however is specifically design for a high-latency, lossy medium and as such can deal with long cables. (S)ATA is the exact opposite - to deliver maximum performance latency needs to be kept to a minimum.

Costwise, ethernet requires the use of magnetics for galvanic seperation. This is important because the longer cables mean you cannot use a common ground between systems. Since the magnetics (basically transformers) are phyisical devices that cannot be implemented on pure silicon, they are expensive. A complete Ethernet PHY will cost a couple of dollars minimum, while a USB chip can be made for a few cents. For DSL, the story is similar.

In summary, it isn't the cable that is necessarily the limiting factor (although they're designed to cost too).

The non technical, and maybe more 'real' answer is that each technology is designed to meet its specific requirements at a specific time. The physical layer is only a small part of the story.
So why does one support longer distances than the other? The answer is 'because they're designed that way'.
It all comes down do 'design to cost / requirements'. The main reason so many different transmission protocols exist if because there is 't one that meet all requirements for everything. As such, companies design new systems that meet all their requirements, but one other. Yes, it would theroetically come up with a single system that covers 90% of everyone's needs, but that then immediately means the companies can't make money on their own proprietary systems. Furthermore, anything you can come up with will be out of date in a few years.

\$\endgroup\$
  • 1
    \$\begingroup\$ RJR, Wikipedia says even USB 2 uses a twisted pair, so certainly USB 3's pairs would also be twisted. And it appears USB cables and cat6 use similar wire gauge, around 24 AWG. \$\endgroup\$ – Porthem Oct 28 '14 at 4:47
  • \$\begingroup\$ SATA Express (NVMe or AHCI over PCIe) was created because SATA was too slow. And latency limits don't explain why Displayport is limited like SATA and USB 3 are. \$\endgroup\$ – Porthem Oct 28 '14 at 4:56
  • \$\begingroup\$ You're right, I'll address the USB twisted pair in the answer. Displayport isn't necessary 'limited' - it is designed to a specification (which I guess Apple now isn't happy with as their new retina iMac display required a higher bandwidth than supported by displayport and thunderbolt. As such, they're can't make a retina external display without inventing a new interface). \$\endgroup\$ – RJR Oct 28 '14 at 5:00
  • 1
    \$\begingroup\$ @RJR - cost would certainly explain it, but then, why does it cost more? mkeith said "ethernet incorporates highly sophisticated signal processing", but what is it? Is it the multiple carrier frequencies, adaptive rates, and noise measurement and cancellation that xDSL uses, or something else? Signal power would explain it, but is ethernet's power really so much higher than the others, to account for its 50x range advantage? Backward compatibility doesn't explain Displayport's or USB 3's (not USB 2's) range limits. High power and signal processing seem the most plausible general explanations. \$\endgroup\$ – Porthem Nov 8 '14 at 23:28
  • 1
    \$\begingroup\$ @Porthem ethernet for example requires the use of magnetics for galvanic seperation. The is important because the longer cables mean you cannot use a common ground between systems. Since the magnetics (basically transformers) are phyisical devices that cannot be implemented on pure silicon, they are expensive. A complete Ethernet PHY will cost a couple of dollars minimum, while a USB chip can be made for a few cents. For DSL, the story is similar. \$\endgroup\$ – RJR Nov 8 '14 at 23:33
0
\$\begingroup\$

As i see from the updated description and collected comments for the question ("Why does Ethernet on UTP have much greater range than other modern protocols?"), the OP is interested in two aspects (and i assume, more much in the second :-) here:

1) common sense, that he defined as

I'm asking in electronics rather than networking since this is a physical link issue, not a networking issue.

2) particular sense, about that he let out in his comment

... it'd be nice to be able to plug in remote webcams (USB) a few tens of meters away for use as cheap security cameras, and to move a noisy computer to a closet 20m away from a display (Displayport).

Please me to start from the second aspect because it seems to me more simple.

20+ m USB link

There is no problem with such a length USB link. In that case, you need to use so called "active extension cable" (one or several stacked in series), like this, for example. It single can cover a span up to 20 m. Yes, in that case you need to power the distant cam separately (i.e. not by the AEC), but i think it's a small additional cost for playing that scenario. And yes, it's expensive (~100 USD/pcs), but the target cams are cheap :-)

For DisplayPort, AECs also exist, try this, for example. They are also expensive (~100 USD/pcs for the 15 m long AEC referenced). But as i understand you need only 1-2 pcs and only once for the next 5-10 noise-free years :-)

Comparing to them, a 20 m Cat5e UTP looks so cheap (~10 USD/pcs) and then magnetically, but an Ethernet enabled cam does not in all. Therefore, you can use the cheap USB cam engaged with a USB to Ethernet Converter (search E-Bay, ~5 USD/pcs), and the last one issue we still cannot overcome here is the need in an extra power path for the cam.

Then, as you can see, your particular problem has a solution.

Why Ethernet can...

RJR in his answer shows several technical reasons i find spurious ("sophistication", "dma-less", "cost") but it's not the case, and secondary --- it's the case.

The primary reason is simple and it is... (ta-da:-) the power necessary to enable the link.

Look up at the numbers:

100BASE-TX: Micrel KS8041 PHY (with xformer)  ....
  for about 200 Mbps (both dirs),  consumes  about 0.33 W,  i.e.  ~3 mW/m @ 100 m span

1000BASE-T: Micrel KS9021 PHY (with xformer)  ....
  for about 2 Gbps (both dirs),    consumes  about 1.12 W,  i.e. ~10 mW/m @ 100 m span

USB 2.0: FTDIChip FT232R IC (self feed only)  ....
  for about 480 Mbps (both dirs),  consumes  about 0.08 W,  i.e. ~15 mW/m @ 5 m span

G.SHDLS: Infinion SOCRATES IC (with hibrid)  ....
  for up to 4 Mbps (both dirs),    consumes  about 2.00 W,  i.e.  ~2 mW/m @ 1000 m span 

How could this be interpreted? I prefer that:

  • if you want a faster speed, you need more power,

  • if you want a longer distance, you need more power too,

  • if you want a wider SNR margin, you need more power again,

BUT:

  • in any way, keep your technology power efficient.

In other words: Why USB wins table-long distances? because it's inefficient to waste 4(12) times more power when possible to work with only 0.08W. And why Ethernet wins building-long distances? because again, it's inefficient to waste 5(2) times more power when possible to work at only 0.25(2.5) of initial speed.

All other reasons, if any, are only and only secondary.

P.S. For that my opinion about RJR's tech reasons' "spuriosity" not to be naked, i'll promise to (try to) describe that informatively as i can if anybody addresses a separate question about it (i would not explain it here because it's far out of the scope of the OP's question).

P.P.S. Also, as i found my previous answer downvoted, i think i had not explain the similarity between 1000BASE-T and xDSL clear enough to be understandable by a side reviewer. Therefore if also anybody asks a separate question about that similarity, i promise to (try to) answer it too.

\$\endgroup\$
-2
\$\begingroup\$

100BASE-T2 and -T4 (not -TX), 1000BASE-T, 10GBASE-T PHYs use coding schemes working like xDSL ones. xDSLs are last mile technologies and they were designed to reach up distances from kilometers and longer.

SATA, PCI-E, USB 3.0 use coding schemes very similar to BASE-X families of Ethernet (100BASE-X, 1000BASE-X, 10GBASE-X) which are much more simple than the listed xDSL-like Ethernet physical sublayers.

Theoretically, SATA, PCI-E, USB-3.0 could be mapped on the listed xDSL-like Ethernet PCS/PMA layers, but... for what? They all are a sort of peripheral (data highways), not of telecom.

\$\endgroup\$
  • \$\begingroup\$ Ok, mkeith says it's because ethernet's signal processing is sophisticated, DoxyLover says 10GBaseT is more expensive (makes sense if the processing is more sophisticated), and alex answered with details about the more sophisticated signal processing. Assuming it's correct, this is exactly the explanation I was looking for. But why was alex's answer downvoted? To answer alex's question (for what?), it'd be nice to be able to plug in remote webcams (USB) a few tens of meters away for use as cheap security cameras, and to move a noisy computer to a closet 20m away from a display (Displayport). \$\endgroup\$ – Porthem Oct 30 '14 at 6:27
  • \$\begingroup\$ (a day later)... no explanation of the downvote on alex's answer, so I'm accepting his answer. \$\endgroup\$ – Porthem Oct 31 '14 at 6:03
  • \$\begingroup\$ This is complete nonsense. The coding scheme hardly bears any relation to distance. Evey modern digital transmission scheme apart form the most trivial has some sort of error correction. The way xDSL achieves its distance is by using multiple frequencies, active noise measurements and adaptation of the modulation rates for each frequency and (for most modern systems using vectoring) noise cancellation techniques. None of this has anything to do with the coding scheme (which is basic trellis encoding). \$\endgroup\$ – RJR Nov 3 '14 at 3:35
  • \$\begingroup\$ Also, DSL and Ethernet do NOT use the same encoding scheme. Ethernet uses PAM (Pulse Amplitude Modulation) or NRZ, while xDSL uses QAM (Quadrature Amplitude Modulation) which takes into account both phase and amplitude of the signal. \$\endgroup\$ – RJR Nov 3 '14 at 3:43
  • \$\begingroup\$ @RJR - SE didn't tell me you added comments; I just noticed them now. I unaccepted alex's answer. If gigabit Ethernet (and higher) over UTP don't use multiple carrier frequencies, adaptive rates, and noise measurement and cancellation, then I still can't find a plausible answer to my question, aside from mkeith's general "highly sophisticated signal processing". \$\endgroup\$ – Porthem Nov 8 '14 at 22:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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