> I have a intel box running Redhat 6.1(2.2.12-20) with 2 ethernet cards
> connected to same subnet, say 10.1.16.0/20. Now I was testing the system for
> network fault-tolerance. And I found something strange,
> which is as follows:
> If I pull out the ethernet cable of "eth0" "eth1" stops working,
> Whereas the second interface i.e. eth1 do replies to the ARP queries
> for both interfaces with its MAC address, but nothing more one can
> do with the second interface.
> Whereas if I keep the cable plugged in for eth0 and pull out the cable
> of eth1, system remains available and starts accepting the packet for
> the other interface also ( packets for other interface).
>
> My question is should'nt both interfaces behave in similar way.
> If you do not know the answer, please tell me some pointers.
Lets say, eth0 is 192.168.0.1 and eth1 is 192.168.0.2.. now if on eth0
arrives an arp-request for 192.168.0.1, eth0 will reply with its
HW-address... the problem is, that the same arp-request is seen by eth1,
too, and eth1 _also_ answeres with its HW-address... usually the second
reply counts...
Now the other machine starts sending the IP-packets to 192.168.0.1 but
using the HW-address of eth1, so if you pull out the cable on eth1 now,
the connection hangs
I've recently posted a "quick dirty hack" which breaks many cases to the
LKM, of course it was rejected it was just to show, how I would expect
the behaviour in this case... then someone pointed me to a new feature
of 2.2.14 which allows hiding interfaces from global arps...
echo 1 > /proc/sys/net/ipv4/conf/all/hidden
echo 1 > /proc/sys/net/ipv4/conf/eth0/hidden
echo 1 > /proc/sys/net/ipv4/conf/eth1/hidden
I haven't tested this yet, give it a try... but you need kernel 2.2.14
for this!
Flo
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
