Network Topologies and Design Patterns
Key Takeaways
- Star topologies centralize endpoints on one switch; the central device is the single point of failure.
- Full mesh of n nodes needs n(n-1)/2 links, so cost and complexity explode; partial mesh balances resilience and cost.
- Spine-leaf uses every-leaf-to-every-spine links with ECMP for predictable two-hop east-west paths and never directly links leaf-to-leaf.
- Three-tier (access, distribution, core) suits large campuses; collapsed core merges distribution and core for smaller sites.
- Hub-and-spoke centralizes branch traffic on a hub (a bottleneck and failure point); point-to-point is one dedicated link.
Reading Topology Questions
Topology items ask how devices interconnect and what happens during growth or a single failure. The right answer follows the dominant traffic pattern and the resilience requirement. Watch for the words "redundant," "east-west," "bottleneck," "single point of failure," and "scale."
| Topology | Meaning | Strength | Weakness |
|---|---|---|---|
| Star | Endpoints to one central switch | Simple, cheap, easy to manage | Central device is a single point of failure |
| Full mesh | Every node to every other node | Maximum path redundancy | n(n-1)/2 links; costly, complex |
| Partial mesh | Some redundant interconnects | Balances cost and resilience | Needs routing design |
| Point-to-point | One direct link, two nodes | Simple dedicated path | Link loss isolates both ends |
| Hub-and-spoke | Branches via a central hub | Centralized control/policy | Hub bottleneck and failure point |
| Spine-leaf | Each leaf to each spine | Predictable two-hop east-west | Needs consistent design + ECMP |
The full-mesh link count is n(n-1)/2: 5 nodes need 10 links, 10 nodes need 45. That math is why full mesh is reserved for small, critical cores and why partial mesh dominates real WANs.
Campus Designs: Three-Tier and Collapsed Core
Large campuses use three hierarchical tiers; smaller sites collapse the top two into one block.
| Layer | Role | Typical devices |
|---|---|---|
| Access | Connects endpoints | Access switches, APs, phones, cameras |
| Distribution | Aggregates access, applies policy | Layer 3 switches, ACLs, routing boundary |
| Core | High-speed transport between blocks | Redundant high-throughput switches |
| Design | Best fit | Exam clue |
|---|---|---|
| Three-tier | Large campus | Distinct access, distribution, core |
| Collapsed core | Small/medium campus | Distribution + core merged |
| Spine-leaf | Data center | East-west traffic, leaf uplinks to every spine |
The distribution layer is where most inter-VLAN routing, first-hop redundancy (FHRP/HSRP/VRRP), and access control lists live. Collapsing the core saves switch cost when there are only a few distribution blocks, but it removes a dedicated fast-transit layer, so it is a small-network choice.
Spine-Leaf Fabrics
In a spine-leaf fabric, servers attach to leaf switches and every leaf uplinks to every spine switch. Leaves do not connect to leaves, and spines do not connect to spines. The result is a deterministic two hops between any pair of servers (leaf-spine-leaf), and equal-cost multipath (ECMP) spreads flows across all spine links. This replaces the old three-tier model for data centers where most traffic is server-to-server.
| Traffic pattern | Definition | Design concern |
|---|---|---|
| North-south | Client/internet to server | Edge/firewall throughput |
| East-west | Server to server, storage replication, microservices | Consistent low-latency fabric |
| Predictable latency | Same hop count for any pair | Spine-leaf delivers two hops |
| Fast growth | Add ports vs add bandwidth | Add leaves for ports, spines for bandwidth |
When the scenario emphasizes heavy east-west traffic and unpredictable aggregation bottlenecks, spine-leaf is the design answer. To add server ports you add leaves; to add fabric bandwidth you add spines.
WAN Patterns and Worked Scenarios
Hub-and-spoke routes every branch (spoke) through a central hub, common when sites share a headquarters or cloud security stack. Point-to-point is one direct circuit between two sites. Mesh WAN gives sites direct paths but multiplies cost and complexity.
| Scenario | Likely pattern |
|---|---|
| Branches send all traffic through HQ | Hub-and-spoke |
| Two data centers on one private circuit | Point-to-point |
| Every branch can reach every other branch directly | Mesh |
| Office endpoints all home to one access switch | Star |
Scenario 1: A data-center app has heavy server-to-server traffic and unpredictable congestion through aggregation switches. Spine-leaf is correct: every leaf has equal uplinks to all spines, ECMP balances load, and any server pair is two hops apart.
Scenario 2: A small campus has two access closets and one redundant pair of Layer 3 switches doing both policy and fast transport. That is a collapsed core, because distribution and core roles are combined rather than separated into three tiers.
Failure-Impact Drill
Expect questions that ask what breaks during a single failure and how to mitigate it. Reason from where the single point of failure sits.
| Failure | What breaks in a simple design | Mitigation |
|---|---|---|
| Star central switch fails | All attached endpoints lose connectivity | Redundant access/uplinks where justified |
| Hub site in hub-and-spoke fails | Spokes lose shared paths and central services | Redundant hubs or backup direct paths |
| One point-to-point link fails | Both endpoints lose that path | Secondary circuit or alternate route |
| One spine fails in spine-leaf | Leaves keep using remaining spines | Multiple spines + ECMP absorbs loss |
| Single distribution switch fails | One block loses routing/policy | FHRP pair (HSRP/VRRP), redundant uplinks |
Key contrast for the exam: a star fails hard at its center, but a well-built spine-leaf degrades gracefully because losing one spine only reduces bandwidth, not connectivity. That difference in blast radius is exactly what topology questions probe, so pair each design with its single-failure behavior as you study.
A data-center design connects every leaf switch to every spine switch and uses equal-cost multipath to support predictable east-west traffic. Which topology is this?
A small campus combines distribution and core functions into one redundant Layer 3 switch pair. Which design is described?
A full-mesh WAN must connect 5 sites with a direct link between every pair. How many links are required?