Bonding and pocket loss

Robin Mordasiewicz robin at mordasiewicz.com
Tue May 23 11:43:24 CDT 2006


On Tue, 23 May 2006, linux-poweredge at lists.us.dell.com wrote:

> Thomas_Chenault at dell.com wrote:
> Dear Thomas,
>
>> Bonding mode 0 (balance-rr) generally requires that all member interfaces are connected to the same Ethernet switch and that the connected switch ports are configured for link aggregation (etherchannel or similar). Has the switch been configured to match the bond?
>>
> Thanks for your answer. The brief answer is: no.
>
> Each of the ethernet interfaces connected to one of the switches.
>    blade 1  ethernet 0  is connected to  switch1  /  ethernet 1  is
> connected to  switch2
>    blade 2  ethernet 0  is connected to  switch1  /  ethernet 1  is
> connected to  switch2
>     ......
>
> Switch1 and switch 2 are interconnected with 3*1GE.  I know I have to
> check the spanning tree settings on the switches.
> Let us assume it is enabled for each of the ethernet interfaces towards
> the blades.
>
> Is there any bonding mode that can cope with our arrangement?
>
> The original idea is/was that if any of the internal blade switch fails
> the services  still can provided.
>

we have bonding working like this across switches. The keyt is that your 
switches must be in a "forced", not negotiated aggregation mode.
The key is "channel-group 13 mode on"

an example on a cisco 3750 as follows.

ie.
interface Port-channel13
  switchport trunk encapsulation dot1q
  switchport trunk allowed vlan 802
  switchport mode trunk
!
interface GigabitEthernet2/0/7
  switchport trunk encapsulation dot1q
  switchport trunk allowed vlan 802
  switchport mode trunk
  channel-group 13 mode on
!
interface GigabitEthernet4/0/7
  switchport trunk encapsulation dot1q
  switchport trunk allowed vlan 802
  switchport mode trunk
  channel-group 13 mode on
!





More information about the Linux-PowerEdge mailing list