FC1 multiple ethernet interface issue

[Scot L. Harris]

On Tue, 2005-02-15 at 14:32, David Benigni wrote:
> > (personally I prefer the netstat -rn output, I guess that makes me
> old
> > school.  :)  )
> 
> trust me, I'm the same way.  I still can't get my fingers to type dig
> instead of nslookup.
> 

That one still bugs me too.  :)  Most of the time I just ignore the
deprecation warning and use nslookup. 
 
> > At this point the problem is not at the point you shutdown eth1.  It
> is
> > when you bring eth1 up and all your routes get pushed out that
> > interface.  It is that point that the system must be looking at the
> > addressing you are using and says the new interface has the same
> subnet
> > as the old one so all traffic can go this way.  Depending on the
> actual
> > IP address on the interface it may decide that based on the highest
> or
> > lowest IP address.  Not sure about that but that appears to be what
> is
> > happening.
> 
> > So if you think this has been working with the two interfaces
> splitting
> > traffic, it probably has not.  Look at the packet stats ifconfig
> gives
> > you for each interface.  Setup some file transfers or something to
> each
> > IP address for testing.  I suspect you will see all outbound traffic
> on
> > eth1 and nothing on eth0, even for local LAN traffic.
> 
> I see what you are saying.  So, I wonder if using GATEWAYDEV=eth0 buys
> me anything when tearing down eth1.
> 
> In the past I have had a server with two nics (different subnets) and
> split traffic between both.  Each had a different default gw though. 
> This example is more insteresting because the default gw are the same.
> 
> Thanks, 
> Dave

All of the systems I have used that had multiple NICs were configured
for different subnets on each NIC.  I even had a case where I had 4 NICs
on an NCR 3550.  I can remember having to play some games to get the DNS
to serve out the addressing I wanted so that devices on each LAN got the
address of the locally connected interface instead of DNS just handing
out addresses round robin fashion for the name of the server.  Got it
all working but at the time it produced some interesting results.  HP's
TCP stack had an interesting "feature" (note the quotes) where if the
response packet came back from a different IP address than what you sent
to the packet would be dropped before being passed up the stack.  Took
some digging to track that one down.  

Typically in your situation people are trying to bond two interfaces
together to improve network performance.  This usually involves a switch
that can support that type of configuration as well as the OS on the
server.

Since both NICs are in the same subnet you are going to end up with a
single default gateway which is going to go out one interface only.  Any
devices on other LANs that go through the gateway will always get
replies from that interface.  

You have asymmetrical routing in the current situation.  Packets will
come in one interface but replies will go out the other one.  

Can you get a sniffer in place to verify that?

You should be able to run ethereal on each interface and look at the
traffic that way.
 
-- 
Scot L. Harris
webid at cfl.rr.com

Don't drink when you drive -- you might hit a bump and spill it.