[j-nsp] EX4100/EX4400 1GE autoneg issues on SFP+ ports

[Tobias Heister]

Hi,

> Bernhard Schmidt via juniper-nsp juniper-nsp at puck.nether.net hat am 10.11.2024 22:15 CET geschrieben:
> Does anyone have any idea how I can beat the Juniper side into doing
> autoneg on 1GE fiber?

This is an ever growing "industry" problem. Most modern chips have lost or are losing the capabilities to support 1G at all or (in these cases) to do specific 1G feature. In this specific case its Autoneg for 1G Fibre connections (along CL37).

Notable chips that have these problems are (most) BCM Trident3 variants (used in e.g. EX4100/EX4400/QFX5120/EX4650 etc.) and some BCM DNX J/Q chips of various generations and probably many more that I am not (yet) aware of. Sometimes only on a subset of their ports, sometimes on all ports.

There are ways to mitigate that problem. 
You could add additional HW on the Switch on PHY/MAC layer to implement these features during switch design. Few vendors do this as of today.
You could use "smart" transceivers (like the Arista Rate Adaptable one mentioned above) that implement the missing feature (autoneg in this case) in the transceiver itself. These were developed pretty much due to experiencing these limitations and needing a way around them.

If you are in the situation to have the HW (e.g. EX4100) that lacks the capability, there is not really anything (to my knowledge) to solve that from the vendor side after the fact (besides using smart transceivers, see above). Although Juniper pulled the stunt to implement a Duplex State engine in the field. using a MACSEC Phy on EX4300 back in the day, they were “lucky” as the additional MACSEC HW had that additional trick up its sleeve 😊

I think it’s quite "funny" that after we had mostly beat the autoneg issues of the late 90s and early 00 years, it now has a revenge on all of us.

As this is not vendor specific problem, but HW/Chip Limit, it hits across vendors. I have been bitten by this on e.g. Arista and Juniper HW multiple times over the last years. With the increasing usage of the lower End BCM T3 in all kinds of switches it becomes much more apparent lately ... 
And yes, it is a pain on e.g. SP/Service Edge where you cannot control what customers connect on their end … and will not easily change that … 

regards
Tobias