|
|
Question : Intel Teaming and NLB / Changing NLB cluster MAC address
|
|
I've torn my hair out over this for two days.
The setup: 3 identical dell servers (PowerEdge 1850s), each with 4 Intel 1000 Pro MT ethernet ports 2 Cisco 500G switches (12 port GigE, Layer 2) Each server has two Switch Fault Tolerant adapter teams set up, one for regular comm and one for NLB The Intel teaming documentation indicates that NLB can only be used in multicast mode
If I disable teaming, the NLB cluster can serve data fine (they're web servers). However I absolutely have to have switch fault tolerance before I put these boxes into production.
If I set the NLB cluster up in multicast mode, NLB assigns a virtual MAC that looks like 01005e7f0128 (the second digit is always odd) However, when I try to enter this MAC into the team, the driver responds that it can only take a MAC that has an even second digit I can go into the registry and force the address in, but the cluster won't serve data outside the local subnet
If I set the NLB cluster up in unicast mode, the cluster nodes won't communicate with each other and I can't ping the other's dedicated IP address. I'm guessing this is a result of the Intel cards not being able to work in unicast mode. I've also tried this setup along with not entering dedicated IP addresses.
So, my first thought is to try and change the cluster virtual MAC to something like 02bfc0a80128 (even second digit). Microsoft documentation indicates that it is possible to override the calculated virtual MAC, but doesn't say how to do it. So, can anyone tell me how to do this? (Just changing the cluster MAC registry key doesn't work, it still actually uses the calculated address.)
Alternatively, can anyone tell me how to get these Intel network cards to work with NLB?
Or, should I just dump the $1,200 worth of network cards and replace them with another brand? If so, what should I replace them with?
Thanks in advance!
|
Answer : Intel Teaming and NLB / Changing NLB cluster MAC address
|
|
|
|
|
|
|