domain controllers and hp network card teaming question

[Peter Berger, Jr.]

can someone help me to resolve a debate we have: my colleage states that domain controllers (in our case Win2003SP2) should "not" have their network cards teamed for high availability (via HP's NIC teaming software).? I've not heard of this and cannot Bing/Google/ITRC anything like this. I'm under the impression that a domain controller "should" have it's 2 gigabit NICs teamed to make sure that directory services are highly available.
any information on this would be great. thanks...

[Peter Berger, Jr.]

I have found some links that state that Microsoft Product Support doesn't like load-balanced teaming options.

What about just simple primary and secondary network cards -- one fails, the other takes over -- no load balancing. Here's the exact text from the HP teaming application:

"Network Fault Tolerance only (NFT) with Preference Order. NFT provides the safety of an additional backup link between the server and hub/switch. NFT is implemented with a Primary adapter and a Secondary (backup) adapter(s). During normal operations, if the Primary adapter fails, the link to the Secondary adapter automatically takes over. The user sets the priority of the adapters relative to each other within the team. The priority of the adapters is an additional consideration when determining the Primary adapter."

Seems plausible?

[Rune J. Winje]

I don't see any problems teaming NICs on a DC.

Choosing just fault tolerance is the most compatible solution.

Choosing a load balancing type should also be safe.

Where most get into trouble with load balancing is when connecting the NICs that are teamed to different switches and running into possible switch firmware/os issues or functionality issues/limitations this added layer of complexity gives. A Load balancing team to the same switch should be very safe and simple however.

Microsoft has some valid reservations with respect to using NIC teaming on private heartbeat (aka internal Cluster communication) NICs used in MS Cluster because NIC teaming heartbeat packages may interfere with cluster heartbeat packages.

Further if using Microsoft's NLB (Network Load Balancing cluster) then for the same reason you are even more likely to run into problems.

The cluster related issues can be adressed if selecting SLB (Switch assisted Load Balancing) for the team (then teaming heartbeats are not sent).

For standalone servers (DC's included) however, I don't think there should be any problems.

If planning for conversion to a virtual machine then keeping it simple without teaming is an advantage because a minimum of hardware dependent features/sw/drivers then need to be removed prior to conversion (...and fault tolerance can then be adressed at the VM level by the DC VM automatically starting on another host if the original host gets problems.)

Finally be aware that Windows 2003 SP2 by default introduces and enables a new feature called Scalable Network Pack. In practice more have seen problems with this new feature (affecting single NICs or teamed NICs alike) than any teaming feature. See:
http://support.microsoft.com/kb/948496



Cheers,
Rune

[Igor Karasik]

We use NFT Teaming on our DC - no problems