I have been reading a lot about modern fiber networking recently and I have a clarifying semi-hypothetical question:

Imagine I have two 3-foot sections of Multi Mode Fiber, and four SFP+ 10G transceiver modules. I have two servers, each with an SFP+ slot, mounted 5 feet apart. I can't connect them with one of my cables, but if I could use both cables I could pull it off. I could use an optical connector of some kind and that would probably work, but I don't have one of those.

Does such a device exist that would accept my two extra SFP+ modules, and provide a direct connection? Like a switch, but without the switching. It doesn't really need to know anything about Ethernet Frames, so I don't need a MAC layer. If the configuration was possible, I could use two PHY ICs back to back blindly passing data. But do I need PHYs? I read the SFF-8418, and the SFI interface is quite bare bones. I also saw an SFP+ module app note, and it looks like the module isn't much more than a laser and conditioning circuits. the SFP+ does no buffering, etc.

So can I just make a board with two SFP+ slots wired directly together? Both modules would need power. There are those I2C pins, which probably need to be used to turn on the laser. What SFF document has information on that protocol? Do SFP+ modules need a clock signal?

Assuming I can build a board with properly impedance matched traces, how close to impossible is this?

If it is possible, what are boxes like this doing? some kind of signal reconditioning for long hauls?


  • 2
    \$\begingroup\$ It seems like buying the optical couplers would be a lot cheaper and easier than what you’re trying to do. \$\endgroup\$ – DoxyLover Dec 28 '17 at 8:33
  • \$\begingroup\$ I know, its a hypothetical. \$\endgroup\$ – Tim Vrakas Dec 29 '17 at 1:31
  • \$\begingroup\$ It may be hypothetical but it has real-world issues. Building a 10G interface is out of the question as the software and emulation tools would cost thousands of US dollars. The optical coupler approach seems the easiest. Basically a gender bender. \$\endgroup\$ – user105652 Dec 29 '17 at 5:00

So can I just make a board with two SFP+ slots wired directly together?

Given you have only ~1.5 m of cable between the transceivers, it will probably work if they are transceivers designed for short distances (SR type, for example). If these are long-distance transceivers (LR or ER types, or single-mode types), the links may be too short and you might overload the receivers. If the link were very long (say, more than 1/4 the reach distance of your transceivers) there would likely be too much jitter in the cascaded link.

Both modules would need power.


There are those I2C pins, which probably need to be used to turn on the laser.

Most SFP+ modules can be pin-strapped to turn on the laser without need for I2C programming. Check the datasheet for your transceivers.

What SFF document has information on that protocol?

The protocol isn't defined by SFF, but by the higher level standard, such as Ethernet or Fibre Channel.

If these are 10 Gb/s links in Ethernet, the PMA (Physical Medium Attachment layer) protocol is most likely 64b/66b encoding. If they're 8 Gb/s Fibre Channel, the protocol is 8b/10b, which is more forgiving than 64b/66b.

Do SFP+ modules need a clock signal?

No, they don't require any clock input.

That said, it would be cheaper, and you would get your link operating faster, if you just go buy a 5 or 10 m cable instead of designing and fabbing an interposer PCB to connect two transceivers back to back.

  • \$\begingroup\$ Thank you, and of course you are correct, an optical solution would be the obvious choice in this example. But I was more interested in understanding the underlying electronics. For anyone wondering, the reason for my wondering had to do with this project koruza.net. They use Fiber trancivers to make janky Free Space Optical links. They use a media converter to copper ethernet, but I would prefer fiber, hence the need for an SFI to SFI bridge. \$\endgroup\$ – Tim Vrakas Dec 29 '17 at 1:46
  • \$\begingroup\$ @TimVrakas, It would seem like with the right optics, you could eliminate their box altogether and just collimate light out of a fiber into their system. But I don't know who makes CWDM bidi modules (with their specified wavelengths) for 1 Gbps. AFAIK that's not an industry standard. \$\endgroup\$ – The Photon Dec 29 '17 at 5:59

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