I just want to state this in advance - I'm not a networking engineer, but I'm trying to learn more about how hyper-v clustering should be built from the networking side.

We have a hyper-v cluster that was setup previously and we're trying to figure out why it was setup the way it was. It currently seems like it's setup with 10gig feeding into the supermicro frame switches, on which multiple VLANs are trunked. Then, in the hyper-v switch manager, those VLANs are broken out into their respective networks, and I believe this is how the hyper-v cluster node gets its network connection.

I don't know a whole lot about these supermicro frame switches, but I'm venturing to guess that I can break out the various VLANs into multiple networks at the hardware level so they appear to each of the nodes as physical adapters, and then those can be assigned appropriately to hyper-v, or I can break them out virtually using Hyper-v and the node gets it's connection from the virtual adapter.

So, is there a preferred way to go about this? My gut feeling is to break it out (if possible) before the nodes so that the nodes see separate adapters (management, hyper-v, client (trunked), storage, etc), and then we assign the respective adapters where they need to go instead of relying on the hyper-v service to break it out. Or, do we want to virtualize it all - is one way preferred or more stable?

I hope I got all the terminology right - I'm trying to learn this stuff so your help and patience is appreciated!!

  • I'm somewhat confused. Are you saying that these nodes have a single 10G trunk feeding into their vswitch stack? That's pretty desirable. Are you trying to move away from that to individual hardware links for isolated networks? That might be less desirable. – Spooler Jul 17 '17 at 19:33
  • Actually you want a bit of both.... If I understand you correctly. You should have virtual nics for your nodes for the VLans it requires access to such as management. However these virtual nics should not then be presented to the VMs even if it is to access the same VLAN. This article might help you out altaro.com/hyper-v/the-hyper-v-virtual-switch-explained-part-1 – Drifter104 Jul 17 '17 at 20:08
  • So, I would venture to guess that I can break out the trunk feeding into the supermicro frame switch in to virtual switches that will then be presented to the cluster nodes (they'd essentially see them as physical switches, I'm guessing), or I present the 10G trunked adapter to the cluster nodes and let hyper-v break it out. Does that make sense? I will check out that link now... – Mike Jul 17 '17 at 20:21
  • this sound right: I can bring a 10G line to the virtual frame, and then if possible, present a virtual NIC to each cluster host for management, storage, live migration, and client VM traffic - client NIC would be the only one that VMs would see, and it would be trunked with VLANs the VMs need to see. Does that sound right reasonable? The reason I ask is because this guide says there should be at least three physical NICs for each cluster host: blogs.technet.microsoft.com/keithmayer/2012/12/12/… – Mike Jul 19 '17 at 18:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.