We operate with a single deployment schema but augment it with blueprint variants for the specific deployment scenarios and use cases: We use a cluster of identical converged compute/network nodes of two types - standard blade nodes and large 2U rack nodes  for very large instances with up to 96 physical cores and 6 TB of RAM -  which can be easily horizontally scaled.  All nodes are connected point-to-point with up to 8 x 40 GbE using LAG (Link Layer Aggregation) for network high-availability.
Nodes are deployed behind a high-availability architecture of cloud controllers and in maximum proximity to the integration point with your existing servers. We use a scalable storage cluster with dedicated storage nodes, based on a storage cluster (S-Ultra 21, ceph or both), which is in itself connected through a cluster network with redundant paths and connects to the compute/network nodes via a storage fabric (RDMA/Infiniband in case of S-Ultra 21) providing redundant paths to storage. The ILO and management ports of all hosts in the private cloud are connected to our management cloud via L2/L3 VPN. The private cloud connects to your existing LANs, VLANs or IP subnets using a single, usually aggregated (LAG) path (VLAN) or optionally a separate L3 subnet. All network equipment in this scenario is part of the private cloud.