0
\$\begingroup\$

I have a use case where I am attempting to establish a unidirectional Ethernet link where a singular fiber is used to transmit data to the receiver. The Rx of the transmit side and the Tx of the receive side are not physically connected.

The proposed configuration is as follows:

•The PHY is a fiber optic 1000BASE-X (LX or SX)

•Autonegotiation and any link monitoring protocols are disabled

•Full duplex configuration

•A second transceiver on the low (transmitting) side is used to provide a connection to the Rx port on the transceiver attempting to transmit unidirectionally (Ethernet spoofing) to set the signal_detect variable in the PCS

•UDP is the transport protocol and ARP is statically configured

So the question is will this configuration allow for successful synchronization at the PHY level for the transmitting side, i.e. will this "Ethernet spoofing" allow the second transceiver to provide the /Comma and /Idle code groups to the first that are required to achieve synchronization? Does synchronization need to be established on the high (receiving) side to successfully receive data?

Alternatively, will COTS SFP transceivers support IEEE 802.3 Clause 66, i.e. setting the mr_unidirectional_enable PCS variable, if the PHY is not of the passive optical network type and an OAM sublayer is not present?

I'm aware of security appliance data diodes however am trying to avoid that implementation.

\$\endgroup\$
  • \$\begingroup\$ At 1000 Mbps, the PCS functions are typically in a separate PHY chip, not inside the SFP module, so I'm not sure the question in your next-to-last paragraph makes any sense. To answer the bigger question it probably helps to know what PHY chip you are using. \$\endgroup\$ – The Photon Apr 6 '17 at 21:01
  • \$\begingroup\$ I'm just trying to get an understanding if commercially available transceivers support Clause 66 of 802.3, as its fairly new and I haven't found much on it, other than how its defined in the spec itself which states the PHY has to be a passive optical network and an OAM sublayer is present. I'm curious if support for Clause 66 is even provided in PHYs that are not of this type. I don't have an exact PHY, but will likely be embedded switches used for industrial controllers on the low side and Cisco hardware on the high side. Working on putting together a test setup to verify this is feasible. \$\endgroup\$ – Quiet Apr 7 '17 at 18:06
  • \$\begingroup\$ The transceiver doesn't support or prevent implementing Clause 66, because it doesn't perform any PCS functions at all. The transceiver is essentially the "PMA" layer. It is transparent to PCS functions aside from providing a "Signal Detect" output for the PHY to use, and not being guaranteed to work if the signal isn't DC-balanced by 8b10b encoding. \$\endgroup\$ – The Photon Apr 7 '17 at 18:29
  • \$\begingroup\$ Now I read Clause 36, the transceiver is just a small part of the PMA layer. It doesn't do any of the functions in 36.1.4.2. AFAIK those all typically happen in the PHY chip. I should have said the transceiver is essentially the PMD layer, not PMA. \$\endgroup\$ – The Photon Apr 7 '17 at 18:37
0
\$\begingroup\$

Receive synchronization should not be required to transmit data. However, in some cases the transmit chain is disabled at some point if there is no signal on the receive side. What are you putting the transceiver in? NIC? Switch? Custom device of some sort? I have some experience with running SFP+ modules unidirectionally for various photonics experiments. Generally, the SFP+ module and PHY are no problem, it's at the MAC layer or higher that you can get trouble if there is no receive signal. The technique I used with standard 10G NICs was to loop the transmit side back to the receive side of the same module with an optical tap, either a 90/10 tap or I think a 98/2 tap. With driver modifications to mask the link status from the operating system, it was generally possible to run without the loopback, at least for the specific NIC that we were using. I presume this should work for your application as well. However, if you have control over the MAC higher layers and you can configure them to ignore the receive side status, then this may not be required. I have had no issues with running SFP+ modules transmit-only with no receive connection out of an FPGA board.

You may just need to do some experimenting; gigabit NICs and SFP transceivers are rather cheap second-hand.

\$\endgroup\$
  • \$\begingroup\$ The low side consists of industrial controllers which will be using embedded switches of some sort (likely Hirschmann, Stratix, or something similar). The interface to the high side will be via core switch (Cisco or Juniper). Do you foresee issues because of the switch-to-switch interface? Appreciate the feedback, as it at least sounds like it is possible. \$\endgroup\$ – Quiet Apr 7 '17 at 17:56
  • \$\begingroup\$ Low side as in the transmit side? \$\endgroup\$ – alex.forencich Apr 7 '17 at 19:07
  • \$\begingroup\$ Yes. Low (transmitting side, unidirectionally) to high (Rx). The purpose of this implementation is to broadcast data from an unclass to classified network. \$\endgroup\$ – Quiet Apr 10 '17 at 11:55

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.