Which transceivers still link on an Aruba 8360-32Y4C when the interface group is forced to 50g
We are bringing up a pair of 8360-32Y4C as leaves and the downlink plan is deliberately mixed: mostly 25G to servers, a couple of 50G ports to a compute chassis, and a few slow copper drops for out-of-band gear. Because the SFP56 downlinks are grouped and the speed is set per group, I want to understand what a group setting actually forbids before I freeze the port map.
- Aruba 8360-32Y4C, AOS-CX 10.15
- SFP56 50G module in the group I set to 50g
- SFP+ 10G DAC and SFP28 25G DAC in other ports of the same group
- J9054D 100 Mbps SFP planned for two OOB drops
With the group forced to 50g the slower modules never come up:
8360(config)# system interface-group 4 speed 50g
8360# show interface transceiver
1/1/13 SFP56 50G up
1/1/14 SFP+ 10G DAC Waiting for link
1/1/15 SFP28 25G DAC Waiting for linkWhat I have already done:
- reseated both DACs and moved them to other ports inside the same group
- checked that the far ends are configured for their native speed
- confirmed the same DACs link on a switch where the group is left at 25g
So: in a group set to 50g, is it really only SFP56 50G modules that will link? Does a 25g group accept everything slower, and does the same hold for a group set to 10g? And the transceiver guide lists the J9054D 100 Mbps SFP for ports 5-32 - has anyone actually run one there?
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 6
The group speed is not a cosmetic label, it is the speed set the ports in that group are allowed to run. With the group at 50g the only thing that will link is an SFP56 50G module; a 10G or 25G module in that group is not an accepted speed and stays down, which is exactly what you saw.
A group at 25g is the permissive setting - 25G and everything slower links in it, so a mixed row of 25G and 10G DACs plus copper is a perfectly normal configuration:
Two caveats before you freeze the map. Ports 1/1/1 to 1/1/4 are the MACsec-capable ports and they play by their own rules, so plan them separately and test them with the exact module you intend to use. And I would not promise you anything about a group at 10g or about the J9054D at 100 Mbps - I have never had one in a grouped port and nobody I asked has confirmed it either. For the slow end the per-port ranges in the official transceiver guide for the 8360 are still the reference. If you need 100M on ports 5-32 for OOB, buy one module and prove it on the bench before you order sixteen.
Two things decide this and your post is missing both. Which ports make up the group - are any of them 1/1/1 to 1/1/4? Those four are the MACsec-capable ones on the 32Y4C and they do not follow the same rules as 5-32, so a test done there tells you nothing about the rest of the box.
And post
show interface transceiverwith the group at 25g instead of 50g. If the 10G and 25G DACs come up there, the modules and the far ends are fine and you are only arguing with the group speed.Group 4 is ports 1/1/13 to 1/1/16, so nothing in the 1-4 range - those are still empty.
I moved the SFP56 into another group that stays at 50g (it is up there) and dropped group 4 to 25g. The picture changes immediately:
So the modules, the DACs and the cabling are not the isse. It is the group speed and nothing else.
For what it is worth, a 25g group really is that permissive. My lab 8360 has one group at 25g and it currently has a 10G DAC, a 1G-BT copper module and a 25G DAC all up at the same time, with 40G and 100G DACs on the QSFP ports as well. None of that upsets the group setting.
Unrelated but worth knowing while you are building these out: we lost two days on an 8325-32C to 8360-32Y4C link with an X242 40G DAC (JH234A) that refused to come up. The 8325 side reported
Module interface errorand the 8360 side sat atWaiting for link, both on GL.10.07.0020. It was a software defect, not the cable - booting the 8325 from its secondary image 10.06.0110 brought the link straight up, and 10.07.0021 fixed it for good. The 8360 needed no changes at all. So if a module that should be supported reports a module error rather than simply no link, look at the release before you blame the optic.Worth internalising the general shape of this, because it is not an Aruba quirk. The port's enabled speed set gates the module, and these boxes rarely fall back on their own.
Same story on Spectrum: on an SN2100 or SN2700 with Onyx 3.10.4100, a 10G DAC behind an Amphenol QSA (624380003) is read fine over I2C but never links while the port is still 100G only.
mlxlink -d /dev/mst/mt52100_pci_cr0 -p 15 -mspells it out -State: Disable,Enabled Link Speed 0x00b00000,Supported Cable Speed 0x00001001, recommendationCable speed not enabled. Put 10G into the enabled speeds and the port goes LinkUp at 10GbE.And on a UCS 6248 a genuine GLC-SX-MMD gets flagged
SFP Validation Failedpurely because the unified port is still at its 10 Gbps default; select 1 Gbps on the port and the module is accepted as an uplink. Different vendors, same question every time: does the port allow the speed the module speaks.Closing the loop on what I can confirm on my own hardware. A group at 50g takes SFP56 50G modules and nothing else; a group at 25g carries 25G and 10G DACs side by side without complaint. That covers the bulk of the port map, so the seervers and the compute chassis are sorted and the 1-4 range stays out of the plan until I can test it properly.
Still open: I have not tried a group at 10g, and I never got a straight answer on the J9054D at 100 Mbps in ports 5-32. One module is on order and it goes on the bench before anything else. Until then the OOB drops stay on the 25g group with copper, and the transceiver guide is what I quote to the change board for the per-port ranges.