On Tue, 01 Apr 2008, Gerrard Geldenhuis wrote:

> eth0 is plugged into a Cisco 2950 called A
> eth2 is plugged into a separate Cisco 2950 called B
<snip>
> Watching /proc/net/bonding/bond0 this shows that it immediately makes
> eth0 active again.
>
> However immediately upon doing that the ping starts to fail for about 30
> seconds and then after 30 seconds starts working again.

This sounds like a spanning tree issue. To test this you can turn on portfast
on the switch by running 'spanning-tree portfast' on the interfaces
that the HP is connected to. My rudimentary understanding of spanning tree
is that it is an inter-switch protocol that provides path negotiation and
protects against things like network loops. The default setting is to
prevent packets from being transmitted over the port while the spanning
tree negotiation takes place. Setting portfast tells the switch to
transmit packets while the negotiation is happening. Be sure to talk to a
network admin before using something like this on a wide scale.

One of the general problems with using the Link Detection method of
bonding is that it assumes that the interface is active as long as there
is a link. Things like spanning tree can make this assumption invalid. If
you don't mind the additional traffic on your network, you might want to
look into the arp_interval and arp_ip_target bonding module options.

Thanks,

Doug

_______________________________________________
rhelv5-list mailing list
rhelv5-list@(protected)
https://www.redhat.com/mailman/listinfo/rhelv5-list