40G_BASE_SR4_QSFP_PLUS not auto-negotiating automdix with new cables

[cttripp1]

This one has everyone in my office puzzled.  We have multiple OM-4 cables connecting a 5930 and a 5940 using the 40G_BASE_SR4_QSFP_PLUS module, configured into an aggregated link.  The cables were damaged thanks to some clutzy handling of raised floor tiles; while they seem to be working properly, we decided to be safe and replace them with new OM-4 cables.

The problem is, whenever we replace one of the old cables with a new one, the link between the two switches goes down and does not come back up.  Even trying to replace the cables one at a time, the two interfaces (one on each switch) do not come up.  We've verified light, and we've verified the OM-4 cables are up to spec.

It does not appear that automdix is working when using a new cable.  Both switches appear to be transmitting on the same side.  There's no way to manually configure mdix that I've found.  But as soon as the old cable is put back in place, everything starts working again.

The only difference between the old and new cables is length.  The old ones are 16 meters, and the new ones are 25 meters.  I cannot see how that would make any difference.

Does anyone have any ideas about this?

[cttripp1]

Forgot to mention: if I take one of the interfaces out of the aggregation, it starts working.  When I put it back in, it stops.  So, maybe not an automdix problem but a LAGG problem.  But, again, I have no idea why.  If it were a problem with LAGG, it shouldn't work regardless of the cables being used.

[cgu]

Not an expert but AFAIK these type of SPF don't support auto-MDX. I would try to manually swap the pair on one side of the cable.

[parnassus]

It's not totally clear to me...are you aggregating more QSFP+ Transceivers (so multiple 40Gbps links) together in a logical link aggregation group (LAG) or - instead - are you using just a single QSFP+ Transceiver splitted into four logical 10Gbps ports (e.g. using using tengig command at the interface level thus working with, example, tengig 1/0/48:1, tengig 1/0/48:2, tengig 1/0/48:3 and tengig 1/0/48:4) and THEN you are aggregating those ones in a logical aggregation group (LAG)?

[cttripp1]

I am aggregating multiple QSFP+ transcievers into a single LAG.  More acurately, my predescessor did.  But we're doing this on all our HPE switches in our data center, and I see no reason why it would stop working with the new cable as it is identical in every way except length (25m for the new one vs. 16m for the old one).  I've verified the cables work when not part of the LAG, and checked they meet the specifications for the QSFP+ transcievers.