2

Support for 10GBase-T RJ45 SFP+ transceivers in modern switches can be a hit and miss. What is the technical limitation as to why some switches don't support them? So far my understanding on the issue is that it takes a lot of compute power for the transceiver to do this translation. Computer power usually equates with more power hungry transceivers and more heat dissipation which could explain the limitations as the transceivers may be breaching the SFP+ power requirements or thermal envelope.

Having said that while support for 10GBase-T RJ45 SFP+ transceivers is limited in switches most of them do support 10GBase-T RJ45 DAC cables which in effect do a similar translation to copper so I am confused as to why DAC cables are supported while RJ45 SFP+ transceivers are not.

Finally while switch manufacturers claim they do not support 10GBase-T RJ45 SFP+ transceivers in their switches we have seen over the last few years how new third party transceiver modules have appeared that seem to work perfectly. Are these safe to use? My educated guess around this is that it seems that these suppliers managed to pack enough compute power in these newer transceivers using newer silicon process nodes which provide more compute, are less power hungry and therefore dissipate less heat. Am I in the right track here?

Improve this question

1 Answer 1

Reset to default
2

In many enterprises, switches and accessories are required to be under active vendor support. That leaves you with officially supported SFP modules and their support in switches.

Technically, using 10GBASE-T modules requires them to transcode the -R PCS code from the SFP cage to the significantly more complex encoding required for twisted-pair cabling - see this question&answer for more details. This and the inherent power requirements for 10GBASE-T generally limit the reach of those ports to usually 30 m.

For marketing or support reasons, vendors often limit your choice of SFP modules to only allow official and supported modules (brand lock-in). While there may be good reasons to do that, it's sometimes used to achieve unreasonable margins.

Practically, you're caught somewhere in between those limitations. You have to evaluate your requirements and your possibilities.

While SFP+ DACs and 10GBASE-T both use copper, their PHYs are vastly different. DACs use dual-simplex twinax cabling, some low-complexity transmitters and receivers, and the basic -R PCS code coming from the chassis, limiting their reach significantly.

10GBASE-T transmitters need to

  1. decode the -R PCS code
  2. split the data stream into four lanes
  3. apply low-density, Reed-Solomon forward error correction code
  4. apply Tomlinson-Harashima precoding
  5. encode using DSQ128 (PCS)
  6. scramble data to remove DC bias
  7. encode using PAM-16 (PMA)
  8. send data through hybrids for echo cancellation

and reverse the process for the receiver. Obviously, the additional processing requires more power, increases latency, has limited reach and is often more expensive than 10GBASE-SR fiber.

Improve this answer
5
  • Thanks for the deep level drill down Zac, I was hoping you would answer :-) The marketing aspect I haven't considered and makes a lot of sense, restricting support for their own products not only reduces support overheads but will increase revenue as large enterprises will want to stick to supported options. On my own personal setup I don't need to be fully supported so I can play with cheaper third party compatible transceivers. My assumption that better process nodes are helping these third party compatible transceivers stands then.
    – GreenLantern22
    Commented Mar 25, 2023 at 21:41
  • I have always been aware that CAT cable 1000BASE-T or over used the four twisted pairs but never bothered to think what happens in the PHY. Thank you for shedding light into this subject. Also I know understand why 10GBASE-T RJ45 switches are way more expensive than SFP+ 10GbE ones. And I have a lot more respect for RJ45 transceivers / 10GbE RJ45 ports. It's not a trivial task to do all the above!
    – GreenLantern22
    Commented Mar 25, 2023 at 21:42
  • One final comment if I may, DAC can be faster than fibre at short distances it appears. Or at least this is whst this manufacturer says... arista.com/assets/data/pdf/Copper-Faster-Than-Fiber-Brief.pdf
    – GreenLantern22
    Commented Mar 25, 2023 at 21:44
  • @GreenLantern22 If some few nanoseconds rock your boat... I prefer DACs for in-rack interconnects because they're cheaper and easier to handle. DACs sometimes fall through when the brand lock-ins collide between vendors. Fiber is for everything >10 m and better compatibility.
    – Zac67 ♦
    Commented Mar 25, 2023 at 22:13
  • Yeah true the difference is really negligible but it worth noting as the general assumption is that fiber is always faster. So yes for in-rack DAC is still fine as pretty muich same speed. Other advantages are that the cable is much more robust and not subject to accidental bends.
    – GreenLantern22
    Commented Mar 25, 2023 at 22:44

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.