QFX5120-48Y: no auto-negotiation fields on 25G et- ports, and media shows FEC74
Standing up 25G uplinks from a QFX5120-48Y towards a second switch and a couple of new servers. Before I hand the segment over I want the link settings written down properly, and I cannot find where auto-negotiation lives on these ports.
- Juniper QFX5120-48Y
- 25G optics in the SFP28 cages
- et- interfaces, nothing configured on them beyond a description and unit 0
show interfaces et-0/0/31 extensive prints no auto-negotiation section at all - no state, no advertised abilities, nothing that tells me whether it is on or off. The media view does give me something:
Speed: 25Gbps
Ethernet FEC Mode: FEC74What I tried:
- hunting for a speed/duplex statement on the et- interface, and finding nothnig that applies
- comparing against a 1G ge- port in the same box, where the autoneg block is printed exactly as I expected
So: is auto-negotiation on by default on a 25G port, is there any speed or duplex knob that is supposed to be set there, and how do I verify the state of it? And if there is nothing being negotiated, what has to match between the two ends for the link to come up?
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
You are not missing a hidden knob - there is nothing to negotiate. Optical 10G and 25G exist in one mode only, full duplex, and they have no negotiation handshake of the kind 1G copper runs, so a speed or duplex statement means nothing on an et- interface and the extensive output has no autoneg state to print. Comparing it with the ge- port is what is misleading you.
What decides whether a 25G link trains is FEC, and it has to be identical at both ends or the link simply stays down. Your side is already telling you what it runs:
So make the far end match, or pin both ends explicitly:
And
show interfaces et-0/0/31 mediais your verification step - it is the place that shows both the speed in use and the active FEC mode. Put that command in the runbook instead of the extensive one for these ports.One trap while you are in the completer: the coherent modes it offers - hgfec, sdfec, sdfec15, sdfec25 - belong to coherent optics and have nothing to do with a 25G client port. That list is what the platform supports across all media, not a menu of choices for the port you are standing on.
That explains the missing block, thanks. The far end is the second switch and it was sitting on a different FEC mode, which I had never thought to look at because the optics on both sides are the same part. Once both ends read FEC74 in the media output the link came up and has stayed up under load.
Runbook now says: check media, not extensive, and record the FEC mode next to the port.
The same rule bites hardest on breakouts. Break a 100G port out into 4x25G and it applies per lane: each of the four has to agree on FEC with whatever sits on its own far end, and the odd one out simply never trains - I have had QSFP-100G-PSM4 and QSFP-100G-XSR4 links with three lanes up and the fourth dark for exactly that reason, which looks like a broken fibre until you check the mode.
Copper deserves a line too. Length changes the rules there, not just the margin: the standard sorts passive assemblies into CA-N, CA-S and CA-L, three classes that differ in how much loss they may burn and in whether error correction is mandatory. Practical effect is that a CAB-Q-4S-100G at the long end may refuse to train unless RS-FEC is running on both sides, while a short one in the next rack came up with it switcched off. If you mix cable lengths in one rack, note the required mode next to the length in the inventory. Third-party passive copper is usually fine as long as it meets the electrical requirements and presents a compliant SFF-8636 EEPROM, but that is the part nobody verifies until a lane stays down.