I'm testing NSX-T in my lab with only TIer0 edge node and I have a confusion regarding the fp-nics:
Regarding the confusion I have, fp-eth0 interface is now on the same vlan as the namagement nic, I gave it a static IP in that VLAN, and everything works fine, meaning, VMs on NSX-T can communicate with the physical world using BGP.
Again, what's the purpose of the interface(s) that connects to the overlay N-VDS, and why can't I choose a portgroup that already represents a segment (logical switch) on that N-VDS swich?
All VMW guides are only focusing on multi-tier deployment, I need to test all scenarios, single tier, and multi tier.
The fp-eth0 interface uses for the overlay Transport zone and the N-VDS is the one usei for the TEP. The Edge itself has its own TEP and it works to use the management VLAN, but you are using that VLAN for overlay tunnels involving the Edge. Best practice is to use a separate VLAN for TEP, remembering that the VLAN for the Edge VM TEP has to be different than the TEP VLAN of the ESXi host.
The current version of NSX doesn’t show VLAN backed Logical Switches when configuring the Edge VM in NSX. You have to manually connect to the desired LS directly through vCenter until next version which will solve this is released or manually deploy the Edge VM and then connect it to NSX.
Thanks a lot for the useful information. I know the TEP (VTEP previously ), but I never understood what does the guide mean by saying "the edge must use a different overlay TEP VLAN than used by ESXi....".
One more thing please, the design guide says that the cluster is a "compute and edge shared cluster", however, to my understanding, it's not possible, or not recommended to have the edge VM on a NSX-T prepared ESXi host, if this is correct, then, should we dedicate or use existing hosts with only traditional virtual switches? but what if the host has plenty of physical nics, can I host it on a host that has N-VDS, as well as VSS, or VDS?. The design guide didn't clarify that point.
You can have a shared cluster for compute, management and edge, but sizing and design has to be considered. It is usually easier and more secure to segregate at least a management+edge cluster separate from compute. Having everything on prepared hosts with only N-VDS just makes it even harder when you have Edge VM, as you have an N-VDS connected to another N-VDS, although it works without any problems.
If you have hosts with multiple NICs I usually like to have 2 for vDS/vSS and 2 for N-VDS. This way things are much simpler.