Author: Graham Bleach Date: To: Hampshire LUG Discussion List Subject: Re: [Hampshire] One box, 2 NICs - but with the same MAC address
On 05/02/07, Bond, Peter <PBond@???> wrote: > Well, I seem to have won the point on the MAC addresses; thanks to everyone for the input.
>
> The IP address issue remains, but if I've read the docs correctly I *should* be able to do what is required using IP aliases. Can't see much about the behaviour if the alias is set to the same address as the first interface - probably need to try tinkering.
>
> DNS isn't in use on these networks (at the moment), so that is unfortunately a non-starter. If the first network "goes down" (interpreted here as loss of link as that's the only indicator I have in this system), all traffic is to be sent from the 2nd interface instead - BUT the machine has to respond to the same IP address on both ports whilst both nets are up.
We could do with some more information about the network setup here.
In your emails on the subject you've said that the two interfaces,
which must have the same IP address are on two physical networks. I'm
struggling to understand why an organisation would run two networks
using the same subnet, so more context would help here. Do you really
have two separate broadcast domains using the same subnet? How is the
routing to these networks handled upstream? Can you give us any hint
as to why it needs to be done this way? In the event of a failure in
the network path, what do the upstream routers do to ensure that the
traffic gets directed to the working network?
In the event you have such a network design, with routers making the
decision about which of the identical subnets the packets end up in,
then you can simply define the IP address on both interfaces (I don't
think it makes any difference whether it's an alias) and then work out
how to remove a broken one in the event of a failure. If you don't do
this, the routing table on the host might send traffic down the
"broken" interface. It's not a design I'm familiar with and doesn't
seem to have much to recommend it.
A common scenario is two use two separate physical switches, trunked
together to provide a single broadcast domain. Each host has two
physical interfaces, one connected to each switch. The OS then uses
some form of interface failover mechanism (e.g. bonding on Linux, IP
multipathing on Solaris) to detect when an interface has failed and
deactivate it, using the alternate network path. This failover process
is almost transparent to most applications. If we didn't have
monitoring we wouldn't even have noticed the switch failures we've had
in the last couple of years.
