Các loại mạng & TopologyNetwork Types & Topologies

Part of the Network Engineer Roadmap knowledge base.

Overview

A computer network is a set of devices (hosts, servers, printers, phones, sensors) connected so they can exchange data. But not all networks are the same. They differ in scale (how far they reach — a desk, a building, a city, a continent), in architecture (who provides services to whom), in topology (how the links are physically and logically arranged), and in the transmission media that carry the bits (copper, fiber, radio).

Understanding these dimensions is foundational for a network engineer because almost every design decision is a trade-off across them:

• Choosing a star topology for a campus floor gives you simple troubleshooting but makes the central switch a single point of failure.
• Choosing single-mode fiber for a building-to-building link buys you distance and bandwidth headroom at higher cost per port.
• Choosing a client-server architecture centralizes control but adds a scaling and availability burden on the server tier.

This note maps out the vocabulary and the trade-offs so that later topics — network devices, routing, switching, redundancy & HA, wireless, and cloud networking — sit on a solid mental model. If you have not yet read how networks work, start there first.

Scale        Architecture        Topology            Media
─────────    ─────────────       ─────────────       ─────────────
PAN          client-server       bus                 twisted-pair copper
LAN / WLAN   peer-to-peer        star                coax
MAN                              ring                fiber (SMF/MMF)
WAN                              mesh (full/partial) wireless / radio
SAN                              tree / hybrid

Fundamentals

Network types by scale

Networks are commonly classified by their geographic reach. The boundaries are not rigid — they describe typical size and the technologies that tend to be used at each scale.

PAN — Personal Area Network

The smallest scale: devices around a single person, usually within a few meters. Think of your phone paired to earbuds, a smartwatch, or a laptop connected to a wireless mouse.

• Typical size: 0–10 m.
• Technologies: Bluetooth / Bluetooth Low Energy (BLE), USB, Zigbee, NFC, Thread.
• Examples: wireless headphones, fitness trackers, tethering a phone, IoT device clusters in a room.

LAN — Local Area Network

A network confined to a single building or site — a home, an office floor, a school. This is the bread and butter of enterprise networking. LANs are privately owned, offer high bandwidth, and have very low latency because the distances are short.

• Typical size: a room to a building/campus (up to a few km).
• Technologies: Ethernet (IEEE 802.3) over twisted-pair or fiber, Wi-Fi for the wireless portion, VLANs to segment.
• Examples: an office network of PCs, printers and servers on switches; a home network behind a router.

WLAN — Wireless LAN

A LAN where the access layer is wireless. Technically a subset of LAN, but called out separately because Wi-Fi introduces its own design concerns: coverage, interference, roaming, capacity per access point. See wireless networking for depth.

• Typical size: same as a LAN — a home, office, or campus.
• Technologies: IEEE 802.11 (Wi-Fi 5/6/6E/7), access points, wireless LAN controllers.
• Examples: office Wi-Fi, campus wireless, guest hotspots.

MAN — Metropolitan Area Network

A network spanning a city or metropolitan area, larger than a LAN but smaller than a WAN. Often used to interconnect multiple LANs belonging to the same organization across a city, or delivered by a carrier/ISP as a metro service.

• Typical size: a few km to ~50 km (city-wide).
• Technologies: Metro Ethernet, fiber rings, MPLS, historically technologies like DQDB/SONET.
• Examples: a university connecting several campuses across a city; a city government linking municipal offices; a cable/ISP metro backbone.

WAN — Wide Area Network

A network spanning large geographic distances — regions, countries, continents. WANs interconnect LANs and MANs. The largest WAN is the Internet itself. Because WAN links cross public/leased infrastructure, they typically have lower bandwidth (relative to LAN), higher latency, and are usually leased from carriers rather than owned.

• Typical size: tens of km to global.
• Technologies: leased lines, MPLS, SD-WAN, VPN over the internet, carrier fiber, satellite.
• Examples: a company connecting branch offices in different countries; the internet; a bank’s national ATM network.

SAN — Storage Area Network

A specialized high-speed network dedicated to block-level storage. A SAN presents remote storage to servers as if it were locally attached disk. It is a distinct network (often physically separate) optimized for low latency and high throughput between servers and storage arrays. Do not confuse it with NAS (Network Attached Storage), which serves files at the file level over a normal LAN.

• Typical size: a data center or a room (though it can be extended across sites for replication).
• Technologies: Fibre Channel (FC), FCoE, iSCSI, NVMe over Fabrics (NVMe-oF), InfiniBand.
• Examples: a virtualization cluster whose VMs boot from a shared storage array; database servers backed by an all-flash array.

Comparison table

Two more you will hear: CAN (Campus Area Network) — several nearby buildings under one organization, effectively between LAN and MAN — and EPN/VPN (Enterprise/Virtual Private Network) — a logical WAN built as an overlay over shared infrastructure (covered in VPN & Tunneling).

Network architectures: client-server vs peer-to-peer

Independent of scale, networks organize how nodes provide and consume services. The two classic models are client-server and peer-to-peer (P2P).

Client-server

Dedicated servers provide resources or services (files, web pages, databases, authentication) and clients request them. Control is centralized.

• Strengths: centralized management, security and backups; easier to scale service capacity by scaling the server tier; consistent, authoritative data.
• Weaknesses: the server is a potential bottleneck and single point of failure; higher upfront cost for server infrastructure; requires administration.
• Examples: the web (browsers ↔ web servers), email (clients ↔ mail servers), Active Directory / LDAP authentication, most enterprise applications, DNS and DHCP.

Peer-to-peer (P2P)

Every node is both client and server — peers share resources directly without a central authority. Control is distributed.

• Strengths: no single point of failure; cheap and self-scaling (each new peer adds capacity); resilient.
• Weaknesses: harder to secure and manage centrally; data consistency and discovery are harder; performance varies with peer availability.
• Examples: BitTorrent, blockchain networks, some VoIP/mesh systems, small workgroup file sharing, WebRTC media between browsers.

Most real systems are hybrids. A video conferencing product may use client-server for signaling and authentication but shift to P2P for media when possible. Cloud-native systems are overwhelmingly client-server at the service level while using P2P-style gossip protocols internally.

Key Concepts

Physical vs logical topology

A topology describes the arrangement of a network. It has two facets:

• Physical topology — how the cables and devices are actually laid out.
• Logical topology — how data actually flows between nodes, regardless of physical layout.

These can differ. Classic Ethernet is physically a star (every host cabled to a central switch) but was historically logically a bus (all hosts on one shared collision domain in the hub era). Modern switched Ethernet is physically a star and logically point-to-point per port.

Bus topology

All devices connect to a single shared backbone cable (the bus). A signal travels the length of the cable; terminators at each end prevent reflection.

[A]   [B]   [C]   [D]
 |     |     |     |
=●=====●=====●=====●=  ← shared backbone (with terminators)

• Pros: cheap, little cabling, easy to add nodes.
• Cons: the backbone is a single point of failure; only one device can transmit at a time (collisions); hard to troubleshoot; performance degrades as nodes increase.
• Where used: legacy 10BASE2/10BASE5 Ethernet, small/temporary setups, some industrial fieldbuses (CAN bus in vehicles). Essentially obsolete in modern LANs.

Star topology

Every device connects to a central node — typically a switch (or a hub in the past). All traffic passes through the center.

      [A]
       |
[B]--[SW]--[D]
       |
      [C]

• Pros: easy to manage and troubleshoot; a single link/node failure does not affect others; easy to add/remove nodes; high performance with a switch (each port its own collision domain).
• Cons: the central device is a single point of failure; cost of the central switch; total ports limited by the switch.
• Where used: the dominant topology for modern LANs / Ethernet and Wi-Fi (APs are the “center” for wireless clients).

Ring topology

Each device connects to exactly two neighbors, forming a closed loop. Data travels around the ring (uni- or bidirectional). Some rings use a token to control who may transmit.

   [A]---[B]
    |     |
   [D]---[C]

• Pros: predictable performance, no collisions with token passing; orderly access; dual rings (e.g., FDDI, SONET) give fast failover.
• Cons: a single break can bring down the ring (unless dual/self-healing); adding/removing a node disrupts the ring; troubleshooting can be tricky.
• Where used: legacy Token Ring and FDDI; still relevant in carrier/metro networks (SONET/SDH rings, Metro Ethernet Ring Protection, RPR) for their fast, deterministic failover.

Mesh topology

Devices are interconnected with many redundant links. In a full mesh, every node connects directly to every other node; in a partial mesh, only some (usually the important) nodes are fully interconnected.

Full mesh (4 nodes)          Partial mesh
   [A]-----[B]                 [A]-----[B]
    | \   / |                   |       |
    |  \ /  |                   |       |
    |  / \  |                  [C]-----[D]
    | /   \ |                        \  |
   [C]-----[D]                        \[E]

• Number of links in a full mesh = n(n−1)/2. For 4 nodes that is 6 links; for 10 nodes, 45 links — cost grows quadratically.
• Pros: maximum redundancy and fault tolerance; multiple paths avoid congestion; no single point of failure.
• Cons: very expensive and complex to cable/manage at scale (full mesh); many ports and links.
• Where used: WAN cores and internet backbones (partial mesh among routers), data center spine-leaf fabrics (a form of partial mesh), wireless mesh networks, and critical links that need redundancy. Partial mesh is the practical sweet spot.

Tree / hierarchical and hybrid topology

A tree (hierarchical) topology arranges stars into levels — a root connecting to branch nodes that connect to leaf stars. This is the basis of the classic three-tier hierarchical design: core → distribution/aggregation → access.

              [Core]
             /      \
      [Dist]          [Dist]
      /    \          /    \
 [Access][Access] [Access][Access]
   |  |    |  |     |  |    |  |
  hosts   hosts    hosts   hosts

A hybrid topology combines two or more topologies (e.g., star-of-stars, star-bus, star-ring). Virtually all real enterprise networks are hybrids.

• Pros: scalable, structured, easy to extend and segment; fault isolation between branches; maps cleanly to core/distribution/access.
• Cons: more complex; upper-tier failure affects everything below it (mitigated with redundancy — see redundancy & HA).
• Where used: essentially every campus/enterprise LAN; the internet as a whole is a hybrid of meshes and trees.

Topology comparison

Transmission media

The physical (or wireless) medium carrying the signal. Choice affects distance, speed, cost, and susceptibility to interference.

Twisted-pair copper

Pairs of insulated copper wires twisted together to cancel electromagnetic interference (EMI). Comes as UTP (unshielded) and STP/FTP (shielded). The workhorse of LAN access cabling. Standard Ethernet copper runs are limited to 100 m per segment.

• Pros: cheap, flexible, easy to terminate, supports Power over Ethernet (PoE).
• Cons: limited distance (100 m), susceptible to EMI, lower max bandwidth than fiber.
• Use cases: desktop/access-layer connections, PoE devices (APs, cameras, phones), short data-center top-of-rack links.

Coaxial cable (coax)

A central copper conductor surrounded by insulation, a shield, and an outer jacket. Older LANs used it; today it is mostly for cable broadband and some RF/antenna runs.

• Pros: good shielding, decent distance for broadband, well-established for cable TV/internet.
• Cons: bulky, less flexible, largely superseded by twisted-pair and fiber for LANs.
• Use cases: cable internet (DOCSIS) last mile, CCTV, antenna feeds.

Fiber optic

Transmits data as light through glass (or plastic) fiber. Immune to EMI, supports enormous bandwidth over long distances, and is the backbone of modern networks. Two families:

• Pros: immune to EMI, huge bandwidth, long distance, secure (hard to tap), lightweight, future-proof.
• Cons: more expensive optics/transceivers, more delicate, needs skilled termination (fusion splicing).
• Use cases: building-to-building and campus backbones, data center spine and long links, MAN/WAN, submarine cables, FTTH.

Wireless / radio

Data carried over radio waves (or, for niche cases, infrared/free-space optics). No cabling, but subject to interference, attenuation, and shared-medium contention.

• Pros: no cabling, mobility, fast to deploy, reaches places wiring cannot.
• Cons: interference and congestion, variable performance, weaker physical security (needs strong encryption), regulated spectrum.

Media selection at a glance

Network access methods (last mile)

The “last mile” is how a customer premises connects to a service provider / the internet. At a high level:

• DSL (Digital Subscriber Line): internet over existing telephone copper (ADSL/VDSL). Cheap and widely available; speed drops sharply with distance from the exchange. Largely being retired in favor of fiber.
• Cable broadband (DOCSIS): internet over the coaxial cable TV network. Shared medium in a neighborhood; asymmetric (faster down than up), though DOCSIS 3.1/4.0 improve this.
• Fiber to the x (FTTx): fiber pushed close to or into the premises. FTTH/FTTP (to the home/premises) delivers the highest and most symmetric speeds; FTTC/FTTN (to the curb/node) runs fiber to a cabinet then copper for the last stretch. Usually delivered via a PON (Passive Optical Network — GPON/XGS-PON) that splits one fiber to many homes.
• Fixed wireless / cellular: 4G/5G, fixed wireless access (FWA), and satellite (increasingly LEO like Starlink) connect sites where wiring is impractical. Common as branch primary in rural areas or as WAN backup.
• Leased lines / dedicated fiber: symmetric, guaranteed-bandwidth circuits (e.g., dedicated internet access, dark fiber, Metro Ethernet) for businesses that need SLAs.

Access technologies also feed SD-WAN designs, where a branch bonds multiple transports (fiber + cellular, say) and steers application traffic across them — see cloud networking and VPN & tunneling.

How these choices show up in real network design

The abstractions above combine into three recurring design patterns:

Campus / enterprise LAN

• Topology: hierarchical tree (core → distribution → access), with redundant links between tiers approximating a partial mesh for HA.
• Media: twisted-pair (Cat6/6a) to the desk and to APs; multi-mode or single-mode fiber for building risers and building-to-building; PoE over copper for APs/phones/cameras.
• Architecture: predominantly client-server (AD, DNS/DHCP, file/print, internal apps).
• Wireless: WLAN overlay (see wireless) for mobility.

Data center

• Topology: spine-leaf (a two-tier Clos fabric — a partial mesh where every leaf connects to every spine) has largely replaced the old three-tier model, because it gives predictable, low, equal-cost latency between any two servers (east-west traffic).
• Media: fiber (MMF for in-row/in-rack short reach, SMF for longer) and DAC/twinax copper for top-of-rack; a separate SAN (Fibre Channel or NVMe-oF) or converged storage network.
• Architecture: client-server services, often microservices with heavy east-west traffic.

Spine-leaf (partial mesh)
   [Spine1]   [Spine2]
    /  |  \   /  |  \
   /   |   \ /   |   \
[Leaf1][Leaf2][Leaf3]   ← every leaf ↔ every spine
  |      |      |
 racks  racks  racks

Branch / WAN

• Topology: hub-and-spoke or partial mesh of sites; increasingly SD-WAN overlays over multiple transports.
• Media / access: whatever the last mile offers — FTTH, DSL, cable, cellular/5G, or leased line — often bonded for resilience.
• Architecture: local services at the branch plus centralized/cloud services reached over the WAN; VPN/tunnels for security.

Best Practices

• Design hierarchically. Use a core/distribution/access (or spine-leaf in the DC) model. It scales, isolates faults, and makes troubleshooting predictable. Avoid flat, sprawling layer-2 domains.
• Match topology to the availability requirement. Add redundancy (partial mesh, dual uplinks, dual-homed devices) where downtime is costly; do not pay for full mesh where partial mesh suffices. Design out single points of failure — see redundancy & HA.
• Right-size the media. Copper (Cat6/6a) for access and short runs; fiber for distance, high bandwidth, EMI-heavy environments, and backbones; single-mode when you need reach or future headroom. Respect the 100 m copper limit.
• Segment by function, not geography alone. Use VLANs and subnets to separate user, voice, IoT, guest, and management traffic. This limits broadcast domains, improves security, and simplifies QoS — see switching and IP addressing & subnetting.
• Plan capacity and oversubscription deliberately. Know your uplink-to-access ratios; leave headroom. East-west heavy workloads favor spine-leaf; north-south heavy ones tolerate hierarchical trees.
• Keep the SAN separate (or properly converged). Storage traffic is latency-sensitive and bursty; isolate it on its own fabric or use lossless/QoS-managed converged Ethernet (DCB) if converging.
• Standardize and label the physical plant. Consistent cabling standards (structured cabling, TIA-568), color coding, and labeling save hours in troubleshooting and make the logical topology match the documented one.
• Choose architecture by control needs. Client-server when you need central management, security, and consistency; P2P/decentralized only when resilience and self-scaling outweigh manageability.
• Document both physical and logical topologies. Keep diagrams current. Many outages are prolonged because the real topology differs from the drawing.
• Secure wireless and last-mile transports. Treat radio and internet transports as untrusted: strong encryption (WPA3), and tunnel branch/WAN traffic — see VPN & tunneling.

References

• Cisco — Networking Basics: What is a network topology?
• Cisco Enterprise Campus Infrastructure / Hierarchical Network Design
• IEEE 802.3 Ethernet standard
• IEEE 802.11 Wireless LAN standard
• TIA-568 structured cabling standards (overview)
• Cloudflare Learning — What is a LAN / WAN?
• MDN — How does the Internet work?
• roadmap.sh — Network Engineer