From: Jay Vosburgh <fubar@us...> - 2003-09-17 20:56:36
>I have a Redhat 9.0 (kernel 2.4.20-8) server that has 2 gigabit LAN
>cards, bcm and intel.
>Bonding.o module is inserted with mode=0 (load balance), miimon=100, and
>It works for a while.
The only thing in your described setup that sounds out of
the ordinary is the multicast=1 ("active") setting. Do you see the
same behavior with multicast=all (2, the default)? The
multicast=active setting is nominally intended for use with the
active-backup mode (number 1), not the balance-rr mode (number 0).
The balance-rr mode is also fairly primitive; you may also
wish to try the balance-tlb or balance-alb modes, which have somewhat
more intelligent load balance algorithms.
>But in about 1 minute, tcpdump reports a lot of arp packages (I don't
>know what they are???);
ARP (Address Resolution Protocol) is the system used to
associate network level Internet Protocol addresses (e.g., 10.0.0.1,
typically themselves associated with host names) with hardware level
network addresses (e.g., 00:01:03:05:07:09). The physical ethernet
devices uses these latter addresses to specify the source and
destination for traffic.
The fact that you see a lot of these suggests that the
system's ARP entry for your destination is timing out, but for some
reason is not being updated (possibly due to the multicast setting
causing the ARP replied to be dropped, but I'm just kind of guessing
>and some commands, ex. ifconfig, route...., hangs.
>Unlinking one of the network cords for one second seems to solve this
That would be bad, and may suggest either a problem with the
multicast=active code, or some sort of conflict between balance-rr
mode and the multicast=active setting.
-Jay Vosburgh, IBM Linux Technology Center, fubar@...