[802.3_100GCU] dudek_01a_0112 Comment

[Daniel Dove <ddove@xxxxxxx>]

Hi Mike,

Since we did not have time for me to share my comment/question in the Task Force meeting today, I figured that perhaps via the reflector would be sufficient.

In your backup slide you state;

Having two phys for similar applications has historically been a poor choice. 

-100Base-T4 was an alternate Phy to 100Base-T operating over existing Cat 3 cable with higher complexity instead of using the newer medium (Cat 5 cable). It was not a success. 

My comment: An observation that neither 100BASE-T4 nor 100BASE-T were NRZ or two-level encoding. Actually, NRZI was attempted for the original FDDI on twisted-pair cabling spec but failed to meet the necessary requirements and demanded an unacceptably higher performance media. You may not recall all of the challenges that 100BASE-TX experienced even with the better cabling, but I would cite the two top reasons for its success as follows;

   1) Customers quickly moved to better-than-CAT5 cabling systems as they discovered the signaling was too challenged on marginally passing links. The silicon issues were resolved with DSP and better geometries, but took years.

   2) Full Duplex Switched Ethernet - 100BASE–T4 did not support this and its now the primary form of Ethernet.

If one extrapolates these points to the NRZ only, vs two-PHY argument, #1 says that some customers might find it too difficult or expensive to build high margin backplanes for years, and since both approaches support full duplex item #2 is unlikely to support your case. Thus, I would argue the comparison is invalid.

-10GBASE-LRM was an alternate Phy to 10GBASE-SR operating over existing OM1 and OM2 with higher complexity instead of over the newer medium (OM3). It has not been a success. 

My comment: In order to achieve a superior reach to LRM, 10GBASE-SR used a very narrow spectral bandwidth that ultimately led to a number of years of very low production, higher costs and movements to create a new standard like USR to get around that challenge.

10GBASE-LRM was hardly and equivalent to PAM4. The receiver complexity required to manage a complete, split-cursor scenario and dynamic variations due to vibration were the things that made it so difficult and ultimately not cost effective. 

For those who are not clear on what I mean, a cable or backplane's attenuation characteristics shift very slowly with temperature, and thus are easily tracked. With OM1 fiber, bent around a corner, and vibration from a nearby piece of machinery, the modal distribution of the fiber could change extremely fast from one extreme to the other. It would be like having a backplane go from 4" to 36" a hundred times a second while maintaining BER. 

Finally, I think a better analogy would be the limited and linear interfaces required for SFP+. Specifically, the linear channel used for copper cables. Both of those interfaces are successful and likely to continue for years to come. Each has its own media requirements, and each has a successful customer need.

Best Regards,

Dan