Build the idea first, then use the next activity to check it.
Shapes networks are built in
Two-tier, three-tier and spine-leaf, and what each shape is trying to make predictable.
What you will be able to do
- Describe the collapsed-core and three-tier hierarchies and when each fits
- Explain what spine-leaf makes predictable that a hierarchy does not
- Distinguish a physical topology from a logical one
- Match a site size to a sensible design
Before this: You should know what a switch and a router each do, and what a broadcast domain is - see "Hosts, switches and routers" and "Broadcast domains and VLANs".
Why it matters: Design questions appear on both exams as recognition rather than calculation: given a description, name the architecture. The reason each shape exists is more memorable than its picture, and it is also what tells you when a shape is being used somewhere it does not belong.
Network designs are not arbitrary. Each recognised shape is an answer to a specific problem - usually about how traffic grows, or about how many places a fault can propagate to. Learning the shapes as answers rather than as diagrams makes them much harder to forget.
Physical against logical
Two networks with identical cabling can behave completely differently, so the first distinction is between how something is wired and how traffic actually flows across it.
| Kind | What it describes | Example |
|---|---|---|
| Physical topology | Where the cables actually run | Every desk cabled back to one cupboard - a physical star |
| Logical topology | How traffic flows and what shares a domain | Those desks split into three VLANs - three logical segments over one star |
The hierarchy, in three layers
The classic campus design separates responsibilities into layers, so that each device has one job and traffic between any two points takes a predictable route.
| Layer | Job | Optimised for |
|---|---|---|
| Access | Connect end devices, apply port-level policy | Port count and cost |
| Distribution | Aggregate access switches, route between VLANs, enforce policy | Policy and aggregation |
| Core | Move traffic between distribution blocks as fast as possible | Throughput and simplicity |
The core deliberately does as little as possible. Filtering, policy and routing decisions belong at the distribution layer, so the core can be simple enough to be reliable and fast. A core doing access control is a recognisable design smell.
Collapsing two layers into one
Most organisations are not large enough to need three layers. A two-tier or collapsed-core design merges distribution and core into one pair of devices, keeping the access layer as it is.
| Design | Suits | Why |
|---|---|---|
| Two-tier / collapsed core | One building, or a few | A separate core adds cost and hops with nothing to aggregate |
| Three-tier | Several buildings or distribution blocks | The core exists to connect distribution blocks - with one block there is nothing to connect |
Spine-leaf
A hierarchy assumes most traffic flows from an end device toward somewhere else - upward and out. Data centres broke that assumption: most traffic is now between servers in the same facility, and it can be between any two of them.
In a spine-leaf fabric, every leaf connects to every spine and no leaf connects to another leaf. Any server is exactly two hops from any other server - one up to a spine, one down to a leaf.
| Property | Three-tier hierarchy | Spine-leaf |
|---|---|---|
| Path length between endpoints | Varies with where they sit | Always the same |
| Traffic assumed | Mostly in and out of the site | Mostly between endpoints inside it |
| Adding capacity | Upgrade a layer | Add another spine |
| Loop management | Spanning tree blocks links | All links active, managed by routing |
Scales and boundaries
| Term | What it means |
|---|---|
| SOHO | Small office or home office - typically one combined router, switch and access point |
| Campus | Several buildings on one site, joined by links you own |
| WAN | Sites joined over distances, using links you generally rent |
| On-premises | Equipment you own, in a building you control |
| Cloud | Capacity you rent, in a facility somebody else operates |
The line that matters most in this list is between links you own and links you rent. On a link you own, bandwidth is a purchase you have already made and latency is a matter of distance. On a rented link, both are contractual, and the WAN is where cost per megabit starts to shape the design rather than following from it.
Hybrid, and why it is the usual answer
Most real organisations run some services on their own equipment and some in a rented facility, joined by a connection between the two. That hybrid arrangement is rarely a considered end state - it is what happens as individual services move at different times.
Its defining characteristic is that the boundary between the two becomes a dependency. A service on-premises that depends on one hosted elsewhere now has the joining link in its failure path, and that link is usually the least redundant part of the whole design.
Terminology
- Collapsed core
- A two-tier design merging the distribution and core layers, suited to a single distribution block.
- Spine-leaf
- A fabric where every leaf connects to every spine, giving an identical path length between any two endpoints.
- Logical topology
- How traffic actually flows and what shares a domain, as distinct from how the cables run.
- SOHO
- Small office or home office - usually one device combining router, switch and wireless access point.
Key takeaways
- Physical topology is the cabling; logical topology is how traffic flows over it.
- Access connects devices, distribution enforces policy, core moves traffic and does little else.
- A core layer exists to join distribution blocks, so one block means two tiers is right.
- Spine-leaf makes path length between any two endpoints identical, which hierarchies cannot.
- Consistent latency, not peak speed, is what a fabric design buys.
- The important boundary in WAN and cloud terms is between links you own and links you rent.
Ready to keep going?
Create a free account to save your progress and take the knowledge check.