How fine-grained should logical networks be?
For example, we separate SQL, AD and other service VMs into separate subnets. Should this be modeled as a single "corporate" logical network with the functional network (AD, SQL etc.) subnets as sites, or as separate logical networks? What's the pros/cons of each approach?
At the moment we're single physical site, single subnet per function (AD, SQL) but we might expand to multi-site in the future, so I don't want to make future upgrades difficult.