Huawei CE6800 100GE uplink to a Profitap XX-3200G never comes up with SR4 optics
Replacing a customer's incumbent aggregation pair with CloudEngine kit, and the single link that refuses to come up is the 100GE feed into their packet broker. Everything else on the switch went in without drama.
- Huawei CloudEngine 6800, 100GE port
- 100GBASE-SR4 QSFP28 modules, Huawei parts, one at each end
- Profitap XX-3200G packet broker on the far side
Both ends see their module. The switch log has nothing but transceiver insert and remove entries from when I was reseating things, no alarms whatsoever, and the interface just sits dwn:
<CE6800> display interface 100GE1/0/1
...
FEC : RS-FECTried so far:
- swapped both modules for spares and cleaned the MPO ends, no change;
- moved the link to a different 100GE port on the switch;
- checked the broker side, it detects its module and reports nothing wrong.
So the optics are recognised on both ends, nothing is complaining, and there is no link. What else has to match between a CloudEngine 100GE port and a third-party box before the link will train?
Also available in: English
العربية
Deutsch
Español
Français
हिन्दी
Bahasa Indonesia
Italiano
日本語
한국어
Nederlands
Polski
Português
Русский
Türkçe
Tiếng Việt
简体中文
Comments 3
This is a FEC mismatch. CloudEngine enables RS-FEC by default on 100GE ports with SR4 optics, the broker has no FEC setting and therefore runs without it, and two ends that disagree on FEC never finish training. Nothing is broken, so nothing is logged, which is why the log only has your insert and remove messages.
Turn it off on the switch, in the interface view:
undo fec modedoes the same thing. The second line is the one that matters. CE is two-stage configuration, and an uncommittedfec mode nonelooks perfectly correct when you read the config back while the port stays exactly as down as before. I have watched a customer case run for an extra day for that reason, with everyone convinced FEC had already been disabled.After the commit,
display interfaceshould show FEC: NONE and the port should train.If the far end ever does gain an FEC setting, the better fix is to enable RS-FEC there and leave the switch on its default, since on 100G you actually want the correction. Between two different vendors I would set FEC explicitly on both sides rather than trust anything to negotiate it.
What does the broker say about FEC on its side, if it says anything at all? You have already posted the interesting half yourself: the switch port is running RS-FEC. Huawei is explicit that both ends of a 100GE link must use the same FEC mode, otherwise the interfaces never come up, and that failure looks exactly like yours, both modules healthy, no alarm, no log, port down.
Most packet brokers and taps I have worked with do not expose an FEC knob at all, which means they run without it and the switch is the side that has to give way. Confirm that first, then change one thing.
Worth adding that the CloudEngine default is not a single value, it depends on the module. QSFP28-100G-LR4 and QSFP28-100G-LR1 run with FEC off per IEEE 802.3, while all other QSFP28 types default to FEC on. So the advice to just disable it is wrong on an LR4 link, where the switch is already off and the mismatch lives on the other end.
Two more rules from the same chapter that have caught me:
So check what you are actually holding with
display transceiver interface 100GE1/0/1 verbosebefore deciding which way to push FEC, and remember the commit either way.