Utilising 2 NICs in one server for failover...

[Guest]

Hi!

I have read a bit hereabouts relating to using dual NICs within one system.
I am still confused, however.

If I have a machine (say an HP DL360) with twin ethernet ports, is it
possible to configure this machine so that if one NIC dies (or a cable is
pulled out / cut, or the switch that it's connected to dies), that the
machine can continue to be accessible via the second NIC?

The machine will be connected to (two) Cisco 11503 switches. I understand
that these boxes can do load balancing and failover, but how do you apply it
to a Windows server environment?

Does it make a difference if the machine is running Windows 2000 Server or
2000 Advanced Server?

I do not want to cluster multiple machines, I merely want the machine to
continue to be available on the LAN should one link fail for some reason.

 

[Phillip Windell]

I have read a bit hereabouts relating to using dual NICs within one system.
I am still confused, however.

If I have a machine (say an HP DL360) with twin ethernet ports, is it
possible to configure this machine so that if one NIC dies (or a cable is
pulled out / cut, or the switch that it's connected to dies), that the
machine can continue to be accessible via the second NIC?

No. It absolutely will not. This is well established,...there is no question
about it and we get about 10 questions a week on this.

175767 - Expected Behavior of Multiple Adapters on Same Network
http://support.microsoft.com/default.aspx?scid=kb;EN-US;175767

There are "third-party" solutions with special Nics ($$$) that come with
special software to perform this. HP is one such vendor.

 

[Ryan Hanisco]

Inside the HP tools you will want to set the NICs to failover -- you will be
able to do this either in the HP tools or the Intel tools if you bought
Intel server NICs.

Make sure that you have them in failover as you end up with problems if you
try to aggregate their bandwidth. If that is a real need, you can do this,
but you would need them to the same switch with trunked ports to provide a
single path.

For failover between two switches, you should be good with the standard
failover mode. It will create a virtual NIC that holds the IP address and
two physical NICs that interface with the switch. It works kind of like
HSRP with two logical and one virtual device.

 

[Ryan Hanisco]

And yes. Phillip is right.

This is NOT default behaviour. This REQUIRES the NIC drivers for this. The
DL360, however, has this for it's server-class NICs.

 

[Ed Horley]

Actually, he is running the CSS 11503 series switches, I am assuming with
the additional copper ports (not a one arm configuration with the gbic to a
core switch) therefore it depends on what mode the CSS is running in.

If you are doing active/active mode with the two CSS units and are doing
interface tracking then you will have to be careful how you set this up.
You have to make sure that your active path for the CSS has the active VIP
of the network cards otherwise it will fail back and forth over and over.
Not a lot of fun.

To be honest, if you are using the CSS architecture you are better off with
duplicate servers providing content or running a cluster configuration
behind the CSS. Having a high available switching infrastructure and a
server with dual homed nics is sort of redundancy gone nuts. Of course
every situation is unique so I have no idea what you are trying to do, so
excuse my opinion if they are out of turn. A quick network diagram of how
you have the CSS's set up would be great. Are you doing OSPF or any routing
on the CSS's? Are you doing active/active or active/backup configurations.
Are you using "groups" at all? All of that effects how you would connect
and run the server. Also, if the server is connected to a switching
architecture below the CSS then things are slightly different, especially if
the switch is L3 enabled. The CSS is one of the most difficult network
devices you can come across so be careful. A wonderful tool to managed the
device once you have deployed it is the HSE (Hosting Solution Engine) from
Cisco, makes your life much easier for folks who have to do all the content
pushes and code releases.

We have deployed the CSS 11503's, 11501's, (with the SSL mods) and 11150's
with the SCA's. They can be difficult boxes to debug with stand alone hosts
and HA cluster configurations. Adding a server with redundant interfaces
will be even more fun debugging!

Buy a sniffer or use Ethereal as it will help you understand the flows that
the CSS builds out.

Regards,
Ed Horley
Microsoft MVP Server-Networking