cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1222
Views
0
Helpful
4
Replies

VM-FEX and Windows NLB cluster

Evgeniy Lotosh
Level 1
Level 1

USC B200 M3 servers, UCS Manager v.2.1, system interconnects 6248UP in Ethernet switching mode, VM-FEX technology is used.

I've created two virtual machines running on Windows Server 2012 host. They both has a single network adapter connected to a certain port profile defined in UCS Manager. The system has assigned them unique MAC addresses, and everything works fine - until I begin to join them to a new NLB cluster.

When NLB cluster is created, all the nodes in the cluster mush share the same MAC address. Windows changes MAC address on its network interface replacing hadrware MAC address with a specific MAC address it generates itself. Usually you need only to mark checkbox "Enable spoofing of MAC address" in virtual NIC properies, and Windows handles the rest. Without VM-FEX technology in action, NLB cluster made of virtual machines works fine.

However, when VM-FEX is enabled, system interconnects do not realize that MAC addresses in VMs are changed. It continues to beleive that VMs still has the initial hardware MAC addresses. As a result, network connectivity for those VMs is disrupted completely.

I cannot connect the VMs to the network in a traditional way (without port profile attached, just with VLAN specified in virtual NIC settings). If I try to specify VLAN ID, the systems says that request is not supported (error 0x80070032).

I do need that NLB cluster up and working, and I has no other Hyper-V servers to move VMs there. Any ideas?

4 Replies 4

Evgeniy Lotosh
Level 1
Level 1

OK, here is quick and dirty solution.

Go to server's service profile in UCS Manager and add another vNIC. Use "Windows" policy. Do not assign Dynamic vNIC connection policy. Specify which VLAN are accessible via that vNIC. Let the profile be re-associated with server, boot it and create another virtual switch binding it to that new vNIC. Do not turn on SRVIO extensions for that switch. Add enother NIC to the VM (do not turn on SRVIO extensions for it)), and now NLB cluster can be created.

However, this is bad solution, because it effectively disables SRVIO for virtual machines. Any idea how to build NLB cluster with SRVIO-enabled VM NICs?

I'm having the same issue running Hyper-V Datacenter Server x64 2012 R2 Preview on UCS 2.0(4d). 

Prefer not to create UCS vNICs for each VLAN.  Is there a documented case / work-around that allows for the Hyper-V virtual switch to perform the VLAN tagging per VM?

In fact, I've been using classic vNIC solutions with Cisco servers for a while, and I don't see any glitches.

Thank you.  I expect this is simply my lack of experience with Hyper-V on UCS.  Have been running VMWare on UCS up to this point (on a vBlock).  UCS 2.1 provides good support for Hyper-V with SR-IOV.  Because we need to stay in the vBlock matrix of software, firmware and drivers at the moment, we're not ready to move to 2.1. 

I've been looking for a good reference for 2.0.  Not sure if I should be building a dynamic NIC profile in UCS, or simply using the static VLAN assignments in the VIC profile and installing NIC teaming drivers in the Hyper-V managers to support VLAN tagging in their Virtual Switch?  Trying these out today.  Any advice on the "classic" solutions would be welcome.  Need to allow guests to communicate between blades, so assuming I'll need those Hyper-V virtual switches to provide "External" access.

Review Cisco Networking for a $25 gift card

Review Cisco Networking for a $25 gift card