Describe the common network topologies and where you still see each one.
Show the answer
Say this first
Bus: every node on one shared cable, old coax Ethernet. Star: every node to a central device, which is every modern switched LAN. Ring: each node to two neighbors, still used in metro and SONET or ERPS rings. Mesh: many paths between nodes; full mesh needs n(n-1)/2 links, so it is used only for small cores and WAN cores. Real networks are hybrids, such as a hierarchical star at the campus and spine-leaf in the data center.
Go deeper
Topologies
| Topology | Strength | Weakness | Where today |
|---|---|---|---|
| Bus | Cheap, simple | One break takes all down, shared collisions | 10BASE2 and 10BASE5 history |
| Star | Easy to add and isolate faults | The center is a single point of failure | Access layer |
| Extended star, hierarchical | Scales in tiers | Depends on the uplinks | Campus core, distribution, access |
| Ring | Two paths with few links | Needs a protocol to block the loop | Metro Ethernet rings (G.8032), SONET |
| Full mesh | Maximum redundancy | n(n-1)/2 links | Small cores, DMVPN spoke to spoke, WAN cores |
| Partial mesh | Redundancy where it matters | Design effort | Most WANs |
| Spine-leaf | Equal hops, ECMP | Many cables | Data centers |
Mesh math
10 sites in full mesh need 10 x 9 / 2 = 45 links. 20 sites need 190. That growth is why hub and spoke plus dynamic spoke-to-spoke tunnels (DMVPN, SD-WAN) took over.
Physical versus logical
The cabling shape and the forwarding shape can differ. A hub was a physical star but a logical bus. Token Ring was a physical star with a logical ring inside the MAU.
They will ask next
Why is spine-leaf better than a three-tier tree in a data center?
Any leaf reaches any other leaf in two hops through any spine, and ECMP uses all spines at once. Latency is predictable and east-west capacity grows by adding spines.
Traps
- Forgetting the n(n-1)/2 formula when asked about full mesh scaling.