Originally posted by: AnthonyWilliams

We have multiple VIO Server running 2.2.0.13-FP24 SP03 with a 6-way 8023ad EtherChannel. We're using "10 Gb Ethernet-SR PCI Express Dual Port Adapter (771000801410b003)" adapters (F/C 5708) at the latest firmware level. When testing the redundancy of the EtherChannel and we remove a main adapter from the EtherChannel from the AIX SMIT panels or the "ethchan_config" command while pinging the VIO Server from an external machine, we lose anywhere from 0-5 ping packets. When we run this test by just downing the network switch ports, we lose 0 ping packets consistently.

Our expectation is that when removing a main adapter from the EtherChannel, we should lose no ping packets, but that's not been the case. Is our expectation wrong or has anyone else experienced this issue?

Thanks.

Anthony Williams

Originally posted by: dukessd

Have you set an alternate mac address for the etherchannel interface?

From memory the etherchannel will use the first physical adapters mac address.

If that adapter fails then the etherchannel will use the next available adapters mac address and this switch over may be causing the lost pings, probably as the switch works out what is going on.

If you set an alternate that is not on any of the adapters you can avoid AIX having to change and also avoid any switch confusion.

Just a guess but hope it helps.

Originally posted by: AnthonyWilliams

Thanks for the suggestion. My EtherChannel has six adapters (ent0, ent1, ent2, ent3, ent4 and ent5) in it and the MAC address that is assigned to the EtherChannel is based on the ent5. I removed ent0 from the EtherChannel while I was pinging the VIOS from another system and lost 4 ping packets. I then added the adapter back and removed ent5 from the EtherChannel. There was a much more pronounced hang of the server and I lost 12 ping packets and the MAC address for the EtherChannel became that based on ent4.

So I can definitely see and understand that since we're not using an alternate MAC address, if we remove the adapter who's MAC address is being used by the EtherChannel, there definitely will be an interruption.

I'm trying to get the IBM Support Center to replicate my scenario, but so far they are pushing back saying it's a network switch problem. However, they won't answer my question on what behavior we should expect.

Thanks for the suggestion.

Originally posted by: j.gann

just a note: it's not recommended to use a number of member adapters other than a power of two in an etherchannel . the frames will not be evenly distributed in this case.

J.