Network Types — by Medium, Distance, and Topology — Computer Networks

Summary#

Networks are classified along three independent axes: the medium that carries the signal (guided vs unguided), the scope in distance and administrative reach (PAN, LAN, MAN, WAN), and the topology that wires the nodes together (bus, ring, star, tree, mesh, hybrid). Any one network sits at a point in this 3-D space. Wi-Fi at home is unguided + LAN + star (one AP). A datacenter spine-leaf fabric is guided + LAN + Clos mesh. The public Internet is mostly guided + WAN + irregular partial mesh.

The classification matters because each axis constrains design. The medium determines bandwidth and latency floors. The scope determines who owns the network and the protocols you can run on it (LANs use Ethernet/Wi-Fi; WANs use BGP-routed IP plus carrier links). The topology determines failure modes, scaling shape, and cost.

Why it matters#

Knowing the type tells you which questions to ask. “WAN issue” means BGP, peering, transit links, carrier SLAs. “LAN issue” means Ethernet, switch fabric, spanning tree, ARP. “PAN issue” means Bluetooth or NFC pairing. Each type has its own debugging vocabulary, tools, and likely root causes.

It also matters for system design. When you place compute in two regions, you’ve turned a LAN problem (intra-DC RPCs) into a WAN problem (cross-region replication). Latency budgets change by orders of magnitude. Failure modes change too — LANs fail cleanly (a switch dies); WANs fail partially (one peering point degrades, traffic shifts).

Topology matters where redundancy meets cost. A pure star has a single root that’s both a bottleneck and a single point of failure. A full mesh of N nodes has N*(N-1)/2 links — fine for 4 routers, infeasible for 4000 servers. Real datacenters land on partial-mesh structures (Clos / fat-tree / dragonfly) that buy redundancy and bisection bandwidth without quadratic link cost.

How it works#

Medium: guided vs unguided#

Guided media confine the signal to a physical conductor.

Twisted pair (copper) — telephone wire, Ethernet (Cat 5e/6/6a/8). Cheap, easy to terminate, distance-limited (~100 m for gigabit Ethernet). Twist cancels crosstalk.
Coaxial cable — cable TV, DOCSIS Internet, legacy 10BASE-2 Ethernet. Better shielding, longer reach than twisted pair, still distance-limited.
Fiber optic — light through glass. Single-mode for long distances (10s of km), multi-mode for short (datacenters). Highest bandwidth, immune to electromagnetic interference, expensive to terminate.

Unguided media broadcast the signal through space.

Radio (Wi-Fi, cellular, Bluetooth) — most common. Different frequency bands have different range/throughput trade-offs (2.4 GHz long range, 5 GHz faster, 60 GHz very fast but line-of-sight).
Infrared — old TV remotes, IrDA. Short range, line-of-sight, mostly displaced by Bluetooth.
Free-space optical / lasers — building-to-building links, some satellite ISLs. Very high bandwidth, weather-sensitive.
Microwave / satellite — long-distance backhaul where fiber isn’t economical.

Scope: PAN, LAN, MAN, WAN#

+-------+-----------------+-------------------+--------------------+
| Type  | Range           | Example           | Typical protocols  |
+-------+-----------------+-------------------+--------------------+
| PAN   | ~10 m           | Bluetooth headset | Bluetooth, NFC, USB|
| LAN   | building/campus | office, home, DC  | Ethernet, Wi-Fi    |
| MAN   | city            | metro fiber ring  | Carrier Ethernet   |
| WAN   | country/global  | the Internet      | MPLS, BGP-routed IP|
+-------+-----------------+-------------------+--------------------+

PAN (Personal Area Network) — one person’s devices. Bluetooth, NFC, USB tethering.
LAN (Local Area Network) — one building or campus. Single administrative domain. Ethernet + Wi-Fi.
MAN (Metropolitan Area Network) — a city or metro area. Carrier-operated, often dark fiber rings. Connects multiple LANs of one enterprise.
WAN (Wide Area Network) — country, continent, global. The public Internet is the largest WAN. Enterprises also build private WANs via leased lines, MPLS, or SD-WAN over commodity Internet.

Two intermediate terms show up in datacenter contexts: DCN (datacenter network — a LAN at very large scale, often Clos-structured) and CAN (campus area network — a multi-building LAN, mostly synonymous with “LAN” in casual usage).

Topology#

Bus:    A --- B --- C --- D --- E      (single shared cable; legacy)

Ring:   A --- B
        |     |
        E     C
        |     |
        D ----+                          (token ring, FDDI, some metro)

Star:   A -- + -- B
             |
             hub                          (Wi-Fi AP, Ethernet switch)
             |
        D -- + -- C

Mesh:   every node connected to every other (full)
        or partial (each node has K>2 links)   (datacenter spine-leaf, BGP)

Tree:   root
       /    \                              (hierarchical: aggregation -> ToR)
      /      \
   sw1       sw2
   /\         /\
  A  B       C  D

Bus — all nodes on one shared cable; only one transmits at a time. Original Ethernet. Failure of the cable kills the whole segment. Dead in practice; switched Ethernet replaced it.
Ring — each node connected to two neighbours, traffic flows in a loop. Token Ring, FDDI, some MAN protection rings. Single break can split the ring unless you have dual rings (FDDI did).
Star — every node connected to a central hub or switch. The dominant LAN topology — a Wi-Fi AP serves as the centre, an Ethernet switch likewise. Central node is a single point of failure but easy to replace.
Mesh (full or partial) — multiple paths between nodes. Datacenter spine-leaf is a partial mesh (every leaf to every spine, no direct leaf-leaf). BGP between Tier-1 ISPs is a near-mesh. Survives single-link failures; cost is O(links^2) for full mesh.
Tree — hierarchical. Legacy three-tier datacenters (access / aggregation / core) and home office routers cascaded behind a gateway. Hot at the root, no redundancy if one branch fails.
Hybrid — almost every real network. A datacenter has a Clos mesh between spine and leaf, with stars at the rack level (servers hang off ToR switches). The Internet is a partial mesh of star/tree LANs.

Variants and trade-offs#

LAN (Ethernet + Wi-Fi) — one administrative domain, low latency (~1 ms Ethernet, ~10 ms Wi-Fi), simple addressing (MAC + IP), single failure boundary. Bandwidth cheap and abundant.

WAN (Internet) — many administrative domains, latency in tens to hundreds of ms, complex multi-AS routing (BGP), partial failures common (one link degrades, others OK). Bandwidth expensive.

Star topology — easy to wire (one central device), easy to debug (look at the centre), cheap to scale (add ports). Centre is SPOF; high blast radius if the AP/switch dies.

Mesh / Clos topology — redundant by construction, no single point of failure, supports high bisection bandwidth. More links, more configuration, requires routing protocols (ECMP, BGP) to spread traffic.

Other axes that matter in practice:

Wired vs wireless reliability — wired LAN is essentially deterministic; wireless LAN suffers retransmissions and contention. Mission-critical industrial control runs wired wherever possible.
Symmetric vs asymmetric bandwidth — datacenter LANs are symmetric. Consumer access networks (cable, DSL) are asymmetric — see Access Networks.
Single-broadcast-domain limits — a LAN’s broadcast domain (ARP, DHCP) tops out around a few thousand hosts before broadcast storms hurt. VLANs partition. Routed boundaries are how you scale beyond.

Why did bus topology disappear?

Three reasons. Hardware: a single coax cable was a single failure point — one bad connector or termination killed every host on the bus. Operational: tracing faults required walking the cable. Scaling: CSMA/CD’s collision domain grew with every host added, and throughput collapsed past ~30% utilisation. Switched star fixed all three at once — switches segment collision domains, twisted-pair cabling fails only one host at a time, and per-port wiring is easy to debug. By 2000, hubs (a star-shaped bus emulator) were also gone; everything became switched.

When this is asked in interviews#

Mostly a vocabulary check. Interviewers expect you to know what LAN, WAN, MAN mean; what the canonical topologies look like; the medium options.

Common follow-ups:

“What topology does your datacenter use?” — almost certainly Clos (spine-leaf). Bonus points if you mention ECMP and the why (linear bisection bandwidth scaling).
“Why don’t we use bus or ring topologies anymore?” — bus fails as a single point; ring needs dual rings for survivability; both lose to switched star.
“When would you use Wi-Fi 6 over wired Ethernet?” — mobility, cost of running cable, density of devices. Never for latency-critical or bandwidth-saturating uses.
“Is a VPN a LAN or a WAN?” — neither cleanly. A VPN is a virtual link that tunnels LAN-like semantics over a WAN. Engineers often call it a “site-to-site link” or a “virtual LAN extension”.

Most common in networking screens and early-round infra interviews; tapers off in later, more specialised rounds.