I had to give an overview of our Cloud OS network topology to our support and engineering guys to help them understand at high levels how it all fitted together.
The below diagram hopefully demonstrates the 3 layers – physical, logical and virtual – and more importantly how the fit together.
We have a dedicated Hyper-V cluster for the fabric management domain. We also have a standalone Hyper-V host that has a VM for DC and DPM. As MS recommend a physical DC and a cluster requires access to a DC to come online (Do not place all DC’s inside the one cluster), we found this was the most cost effective solution without compromising on availability. And yes, our DPM is virtualised which makes for a great backup solution. Some handy info can be found here..
Hope this helps give you a high level view on some of the complexity around networking in a NVGRE environment DC.