We have been running an ESX 4.0 cluster for some time, and have been running IPv6 on many of the guests for a year or so.
A few months ago, we built a new cluster with updated hardware running 4.1 and migrated the vms over to it. The machines that migrated function properly.
A couple of months ago, we created a new template vm for Centos 6 to base new vms on. These new machines, as it turned out, do not function properly with IPv6. tshark shows that Neighbor Discovery queries are not getting responses, though with promiscous mode enabled for the vm, it *does* see the ND conversations of other machines. If two of the vms are on the same physical host, they *can* talk to each other however.
In testing, I tried creating Centos 5 vms, and found the same problem. Creating a vm on an esxi 4.0 machine functions properly there, but not after migrating to the new cluster. Creating a centos 6/64bit vm on one of the old cluster machines exhibits the same failure, which is *really* puzzling, since it clearly used to work.
All are vm version 7 machines.
Given the failure of new vms on the old cluster host to work, it makes me wonder if somehow we've changed the way we create the vms without realizing it, but there really isn't much to tweak...
I'm not quite sure where to look now, but will try a few more combinations and see what that shows...