2950 to 3750 etherchannel duplex issue

[Marius Gunnerud]

I can't find an explination for this issue. I am not having the issue, but I heard about it and it got me curious. a 2950 switch connected to a 3750 switch through port fa1/0/1 and 1/0/2 on both switches. The ports are configured correctly for trunk and etherchannel. Speed and duplex are hard coded to 100 full duplex on the 2950 physical interfaces. The issue comes when hard coding speed and duplex on the 3750. Once hard coded one of the links in the etherchannel goes down while the other remains up. However, once the speed and duplex are removed from the 3750 both links in the etherchannel come up (speed and duplex are configured to 100 full duplex on both switches when testing this).

Has anyone come across this before? Anyone have an explanation for it?

[Peter Paluch]

Hello Marius,

I have not personally heard nor seen this issue, and it is a bit surprising to me. I can test it using a 2950 and a 3560 tomorrow, though, and see whether I am able to reproduce it. Do you perhaps know what was the mode of the EtherChannel? (LACP, PAgP, On)

What I can conclude, however, is that it is generally a bad idea to hardcode the speed and duplex. For once, it disallows a device to properly adapt to the capabilities of the neighboring device. The second issue is more involved, however: on older Catalysts, the autonegotiation was completely deactivated when you hardcoded both the duplex and speed, instead of advertising just the hardcoded speed/duplex combination. The neighboring device was usually able to detect the speed basing on the line coding scheme but the duplex was not detectable, thereby falling back to half duplex. This usually led to duplex mismatches. I would hypothesize that either the problem with detecting the opposite party's duplex/speed settings, or a potential duplex mismatch could account for this problem.

Nevertheless, this is just a speculation, and I have no clear explanation for this. Let's see if I can dig something out tomorrow in the lab.

Best regards,

Peter

[glen.grant · Accepted solution]

   That's because on the side you have that goes down you probably have a straight thru cable instead of a crossover and when you hardcode the speed that breaks the auto mdix feature on the 3750 . Auto mdix requires the ports to be set as auto .  There is no reason you have to hardcode the devices , it will work fine with everything set as auto , otherwise try a crossover on the side that goes down.

[Peter Paluch]

Hi Glen,

The idea with the hardcoded speed/duplex deactivating the auto-MDIX is brilliant! Thanks for joining the discussion!

Best regards,

Peter

[Marius Gunnerud]

Sorry for the extremely late reply. I never found out if the setup had a straight-through cable connected but it does make sense.

Thanks for the replies,

Regards,

Marius