I am trying to understand the relationship between VXLAN subnets, clusters and distributed switches.
For example if we want to push Layer 3 down to the ToR switches and have a /24 subnet per rack and we have 20 racks in total, my understanding is that would mean 20 different VXLAN subnets.
In the NSX web GUI we set VXLAN subnets (or “Pools”) at the cluster level. So am I right in thinking 1 cluster = VXLAN 1 subnet?
I would also like to understand the relationship to VDS.
In the VMware hands on lab example they use a separate VDS per cluster. (http://docs.hol.vmware.com/HOL-2014/hol-sdc-1403_html_en/)
In other examples I see a single VDS is used across both compute and management clusters. (http://roie9876.wordpress.com/2014/04/15/nsx-home-lab-part-1/)
So to summarise, if I want to use a different VXLAN subnets per rack, do I need to create more clusters? And should I be using a single VDS for the whole DC or one per cluster? Does it matter?
(Experts correct me if I'm wrong!) Regarding the VDS - something that is quite key is that each VDS can only support a single portgroup for VXLAN transport with a single VLAN ID. In theory you could use a single VDS across multiple racks in your example but you would have to create the same VLAN ID for each and back it with a different subnet at the ToR. This design requires some special treatment (essentially using DHCP for VTEP address assignment versus IP Pools). You can have multiple clusters with their VTEPs in the same subnet so it's not a one-to-one mapping.
Another way to do it might be to create a separate VDS for each VXLAN transport network/rack. You could then add all of these into a single Transport Zone and stretch a L2 Logical Switch across the L3 boundaries and portgroups would be created across all the VDS's (granted 20 VDS seems like a lot).