Network architectures

Campus designs describe how access, distribution and core functions are collapsed. In a three-tier design each layer is a distinct device tier: access connects endpoints, distribution aggregates and enforces policy/routing, and the core moves traffic fast between distribution blocks. A two-tier (collapsed core) merges distribution and core into one tier — common in smaller campuses where a separate core adds cost without value.

Spine-leaf is the dominant data-center fabric: every leaf connects to every spine, so any two leaves are exactly one spine hop apart. Latency is predictable and bandwidth scales horizontally — add a spine for more paths, add a leaf for more ports. It is a Layer 3-oriented fabric where loops are avoided by routed links rather than STP blocking.

A WAN connects sites over service-provider transport — MPLS VPNs, dedicated circuits, or encrypted tunnels over broadband internet. SOHO designs collapse everything into one or two devices: an integrated router with a few switch ports, a wireless radio and a firewall feature set. Redundancy and segmentation are minimal by necessity.

On-premises means you own and operate the hardware in your facilities. Cloud moves workloads to a provider's shared infrastructure (IaaS), platforms (PaaS) or finished applications (SaaS); the network's job shifts to secure, reliable connectivity to the provider — VPN or private interconnect — plus identity-aware access. Hybrid designs keep both.

Common mistakes

• Treating spine-leaf as a campus access design — it targets data-center/fabric traffic patterns.
• Assuming three-tier is always superior — collapsed core is often the right size.
• Forgetting that 'cloud' still needs WAN connectivity and security design.
• Confusing the WAN (inter-site transport) with the internet service itself — MPLS is a WAN without internet exposure.