1

I'm running CentOS 6.5. I have 3 servers with quad 1Gb nics. I'm using bonding mode 6 (tried 0 as well) running through a cisco 4948 switch. When only two network cables are attached, no issues. If more than two are connected, the network randomly drops packets, and we are randomly unable to connect to the server. I was fortunately able to determine that the problem went away when one or two interfaces are connected.

ifcfg-bond0:

DEVICE=bond0
ONBOOT=yes
IPADDR=192.168.1.10
NETMASK=255.255.255.0
NETWORK=192.168.1.0
GATEWAY=192.168.1.1
USERCTL=no
BOOTPROTO=no
BONDING_OPTS="mode=6 miimon=100"

ifcfg-eth0 through 3:

DEVICE=eth0
ONBOOT=yes
USERCTL=no
MASTER=bond0
SLAVE=yes
BOOTPROTO=no

bonding.conf

alias netdev-bond0 bonding
Brian Lovett
  • 197
  • 2
  • 11
  • Have you setup `Etherchannel` for the 3 ports at Cisco switch side ? – krisFR Feb 05 '14 at 18:37
  • I have not. It was my understanding that mode 6 doesn't require any switch specific config changes. – Brian Lovett Feb 05 '14 at 18:51
  • After a quick check it seems you are right that mode 6 seems to not require Etherchannel. I cannot see your bonding mode definition in your question. Can you update it with this infos ? – krisFR Feb 05 '14 at 19:27
  • Sorry about that, I updated the config above to reflect mode 6. – Brian Lovett Feb 05 '14 at 19:43

1 Answers1

1

Sounds like some of your switchports are not in the correct VLAN or suffering some other logical error which separates those switchports from the working switchports.

Try each slave individually, either as the only member of a bond, or without the bond at all.

I expect some slaves will work, and some won't.

(For what it's worth, I can't stand Mode 5 and 6, they are awful hacks and won't work when VMs are bridged onto them. Make friends with your network team, get them to configure an EtherChannel and use Mode 2, or configure LACP and use Mode 4.)

suprjami
  • 3,536
  • 21
  • 29