Today's discussion of
800G FR4-500 baseline was a wonderful example of the breakdown in the utility of the 802.3 SMF Channel Model. For decades, we have used the same worst case min/max manufacturing ITU-T fiber specs for the 802.3 SMF channel model. To give context to the copper track, this is like being told that you still have to use FR4 for PCB material. This didn't matter at lower rates, and the excess penalties just represent additional link margin, or usable loss budget at shorter reaches. Since such fiber is hard to get, it is mostly not used in production testing, which works out great because it is not found in modern data centers.
Unfortunately this charade has bred complacency, which has crept into development. Engineers know the 802.3 SMF channel model is not real so they also ignore it. In the original
800G FR4-500 baseline, the dispersion was 4x the correct value through a cut and paste error, indicative of an area not in need of scrutiny. The dispersion penalty limit was also just copied, and there was no measurement of the dispersion penalty.
Let's look at a nice set of measurements for 400G optics.
For 10km 100G/lane CWDM4, 2.2 and 0.7 dB penalties are reported for EML and MZI, respectively. Scaling this to 500m, 200G/lane CWDM4, we get 0.44 and 0.14 dB EML and MZI penalties. If this decades old 802.3 SMF Channel model was real, that means the proposed 3.1 and 3.4 TEQC and TDEQC are not met. No one is worried and with a smile and a wink the values put into the dispersion row are safely ignored.
It is critical that 802.3 take charge of SMF Channel Models and use statistical distributions of actual fiber in data centers, that can then be taken seriously by everyone, including manufacturing and engineering. However, if we choose to keep the decades old values, or adopt the almost identical and equally irrelevant to Ethernet values being proposed in this meeting, then we should be prepared to start impacting the feasibility and cost of high volume optics. Scaling the Schube values to 2km, 200G/lane CWDM4, we get 1.8dB EML penalty. That's significantly worse than in
the Nov. proposed 800G-FR4 baseline. Starting with 200G/lane, we can't have cost effective optics and eat our charde too.
Chris