Allow me to preface this discussion by saying that this is chiefly a Windows configuration issue, but following their guides has yielded a broken configruration and I am curious as to whether there is something about the ESX virtual networking stack that requires an alternate configuration.
As far as I can tell, best practices for doing Unicast NLB between Windows 2k3 hosts calls for using two network adapters. One for host management, the other for the cluster.
So far, I can only get this to work if I assign a "Dedicated IP" to the cluster adapter, in addition to the Cluster IP. If I do not specify this, I cannot talk to peers in the cluster, even though there is another NIC with a static IP not bound to NLB in each host.
For example, I think it should work like this, with the cluster IP being 192.168.1.100:
-
Host 1
Management NIC: 192.168.1.10
Cluster NIC: 192.168.1.100
Host 2
Management NIC: 192.168.1.11
Cluster NIC: 192.168.1.100
-
I configured it this way using NLB Manager. I did not assign a dedicated IP to the cluster adapter when adding a host to the cluster. However, with the above configuration, I cannot ping Host 1's mangement IP from Host 2, or vice versa, or add the second host to the cluster.
What does seem to work is assigning an additional static IP to the cluster adapter, via the "Dedicated IP" field, when you add the host to the cluster in NLB Manager, yielding this:
-
Host 1
Management NIC: 192.168.1.10
Cluster NIC: 192.168.1.11, 192.168.1.100
Host 2
Management NIC: 192.168.1.12
Cluster NIC: 192.168.1.13, 192.168.1.100
So, does Unicast NLB require 2 Static IPs per host plus the static IP for the cluster, or am I misconfiguring something?
