Reference
Network Architecture
Three-tier (Core/Distribution/Access) and Spine-Leaf network design reference
Three-Tier Hierarchical Model (Campus / Enterprise)
The Cisco three-tier hierarchical model divides a campus or enterprise network into three distinct layers — Core, Distribution, and Access — each with a specific role. This separation makes the network easier to design, manage, troubleshoot, and scale.
The high-speed backbone. Provides fast, reliable transport between Distribution blocks. No policy, no filtering here — just raw forwarding speed.
The brains of the network. Aggregates Access switches, enforces policy, provides inter-VLAN routing, and is the boundary between L2 (Access) and L3 (Core).
Where end devices connect. Provides port-level security, VLAN assignment, PoE, and QoS marking at the edge. Primarily L2 switching.
Three-Tier Diagram
+------------------------------------------+
| CORE LAYER (L3) |
| [Core-SW1] ============ [Core-SW2] | <-- Full-mesh, 10G/40G/100G
+--------+-----------------------+---------+
| |
+-------------+-------+ +---------+---------------+
| DISTRIBUTION (L3) | | DISTRIBUTION (L3) |
| [Dist-SW1] | | [Dist-SW3] | <-- ACLs, QoS, HSRP
| [Dist-SW2] | | [Dist-SW4] | Inter-VLAN routing
+--+-----+------+-----+ +------+-----+------+-----+
| | | | | |
[Acc1] [Acc2] [Acc3] [Acc4] [Acc5] [Acc6]
ACCESS LAYER (L2) ACCESS LAYER (L2)
PCs, Phones, APs PCs, Phones, APs
Spine-Leaf Architecture (Data Center / Modern Campus)
Spine-Leaf is a two-tier, flat topology designed for east-west traffic (server-to-server). Every Leaf connects to every Spine in a full partial-mesh — resulting in a consistent, low-latency 2-hop path between any two endpoints.
Pure routing backbone. Spine switches only connect to Leaf switches — never to servers or end devices. Provides equal-cost multipath (ECMP) forwarding across all uplinks.
Directly connects servers, storage, firewalls, and load balancers. Every Leaf connects to every Spine. Acts as a VTEP in VXLAN overlay designs.
Spine-Leaf Diagram
SPINE LAYER
[Spine-1] [Spine-2] [Spine-3]
/ | \ / | \ / | \
/ | \ / | \ / | \
[Leaf-1][Leaf-2][Leaf-3][Leaf-4] <-- LEAF LAYER (ToR)
| | | |
[Svr] [Svr] [Svr] [Svr] <-- Servers / Storage / Endpoints
Every Leaf connects to EVERY Spine (full partial-mesh)
Server A to Server B = always 2 hops: Leaf -> Spine -> Leaf
Architecture Comparison
| Feature | Three-Tier (Core/Dist/Access) | Spine-Leaf |
|---|---|---|
| Layers | 3 (or 2 collapsed) | 2 (Spine + Leaf) |
| Primary Use | Campus, enterprise LAN | Data center, modern campus |
| Traffic Pattern | North-South (client ↔ server) | East-West (server ↔ server) |
| Latency | Variable (3–5+ hops) | Consistent (always 2 hops) |
| Scalability | STP-limited in L2; good in L3 | Linear — add leaf or spine switches |
| Redundancy | HSRP/VRRP at Distribution | ECMP across all Spine paths |
| Routing Protocol | OSPF / EIGRP at Distribution+ | BGP (eBGP) or OSPF underlay |
| Overlay | VLANs (802.1Q) typically | VXLAN/EVPN for L2 over L3 |
| Complexity | Moderate — well understood | Higher — requires BGP/VXLAN knowledge |
| Cost | Lower for campus deployments | Higher initial investment |
| STP Dependency | Yes (at L2 access layer) | Minimal or none (L3 to leaf) |
When to Use Each
What is a Network Fabric?
A network fabric is the complete, interconnected switching and routing infrastructure that carries traffic between all devices in a network domain — think of it as the underlying “weave” of switches, links, and protocols that forms the transport substrate. The term is most commonly used in data center and modern campus designs to describe the entire forwarding plane as a unified system rather than individual boxes.
In practice, “fabric” often refers to a Spine-Leaf topology running an IP underlay (OSPF or eBGP) with a VXLAN/EVPN overlay — but the word can apply equally to a three-tier campus backbone or an SD-Access fabric.
Underlay vs. Overlay
Every modern fabric has two planes working together:
| Plane | What It Is | Protocols | Job |
|---|---|---|---|
| Underlay | The physical IP network — routers, switches, and links | OSPF, IS-IS, eBGP | Provide reachability between all fabric nodes (VTEPs, spines, leaves) |
| Overlay | The virtual network tunneled on top of the underlay | VXLAN, EVPN, MPLS, GRE | Carry tenant/VLAN traffic across the fabric regardless of physical location |
Common Fabric Types
| Fabric Type | Topology | Underlay | Overlay | Best For |
|---|---|---|---|---|
| Spine-Leaf / IP Fabric | Every leaf connects to every spine | eBGP or OSPF | VXLAN + EVPN | Data centers, modern campus east-west traffic |
| Three-Tier Campus Fabric | Core → Distribution → Access | OSPF / EIGRP | VLANs (802.1Q) | Traditional campus, north-south client-server |
| SD-Access Fabric | Cisco DNA Center managed, LISP + VXLAN | IS-IS underlay | VXLAN + LISP | Enterprise campus with policy-based segmentation |
| SD-WAN Fabric | Hub-spoke or full-mesh over WAN transport | Internet / MPLS | IPsec tunnels | Branch office connectivity across WAN |
Key Fabric Terminology
| Term | Definition |
|---|---|
| Fabric Edge | Where end devices connect to the fabric — a Leaf switch (spine-leaf) or Access switch (three-tier). The first point of policy enforcement. |
| Fabric Border | Where the fabric connects to external networks — ISP, WAN, or another fabric domain. Handles route leaking and external reachability. |
| Fabric Control Plane | The protocol managing host location and reachability across the fabric (e.g., EVPN distributes MAC/IP bindings so leaves know where endpoints are). |
| VTEP | VXLAN Tunnel Endpoint — the fabric node (leaf switch or server NIC) that encapsulates/de-encapsulates VXLAN packets at the fabric edge. |
| VNI | VXLAN Network Identifier — the 24-bit segment ID in the overlay, equivalent to a VLAN but with 16 million possible values. |
| Anycast Gateway | All leaf switches share the same default gateway IP/MAC for a VNI/VLAN, enabling hosts to move across leaves without re-ARPing. |
| BUM Traffic | Broadcast, Unknown unicast, Multicast — traffic that must be flooded across the fabric; handled by EVPN (preferred) or IP multicast in VXLAN fabrics. |
Design Best Practices
| Layer / Element | Best Practice |
|---|---|
| Core | Keep it simple — route fast, no ACLs or complex policy. Redundant links minimum. |
| Distribution | Deploy in pairs with HSRP/VRRP as default gateway. Apply ACLs and QoS here, not at Access. |
| Access | Enable BPDU Guard, PortFast, port-security, DHCP Snooping, and 802.1X on all user ports. |
| Spine | Use eBGP between Spine and Leaf with unique ASNs per leaf for simple, scalable policy. |
| Leaf (ToR) | Each leaf is a broadcast domain boundary. Use VXLAN/EVPN to extend L2 across the fabric. |
| Links | Use routed point-to-point /30 or /31 links between tiers. Avoid L2 trunks in spine-leaf. |
| Redundancy | Minimum 2 Spine switches for HA. Dual-home all Leaf switches to both Spines. |
| Oversubscription | Target 3:1 at Access, 2:1 at Distribution/Leaf, 1:1 at Core/Spine for best performance. |