Network Architectures
A topology says how devices can be connected; an architecture says how they should be connected for a given scale, budget, and goal. This lesson tours the campus models (two-tier and three-tier), the data-centre spine-leaf fabric, the WAN connection shapes, the SOHO all-in-one, and the split between on-premises and cloud.
Why architecture matters
The word topology describes how nodes can be arranged — star, mesh, ring, and so on (Lesson 12 covers these in detail). An architecture goes further: it is a proven design pattern that combines topologies, layers, and roles so that the network stays fast, scalable, and manageable as it grows from tens of devices to tens of thousands.
The central idea in almost every campus and enterprise design is hierarchy. If every device connected directly to every other device, the number of links (and the cost, and the configuration work) would explode, and changes in one corner would ripple everywhere. Instead, engineers group devices by role into layers, and each layer talks mostly to the layer above and below it.
Scalability
Add a new access switch without redesigning the core.
Manageability
Policy lives in one consistent place — the distribution layer.
Availability
Redundant paths at each layer survive a single device or link failure.
The three-tier campus model
The classic enterprise campus design uses three layers. Each has a distinct job, and the exam expects you to know which job belongs to which layer.
| Layer | Job | Typical devices |
|---|---|---|
| Access | Connects end devices (PCs, phones, access points, printers) to the network. Enforces port-level policy: VLAN assignment, port security, Power over Ethernet, and access lists. | Layer 2 access switches |
| Distribution | Aggregates the access switches and is the layer of policy: inter-VLAN routing, filtering, QoS, route summarisation, and redundancy between access and core. | Layer 3 switches / routers |
| Core | The high-speed backbone that moves large volumes of traffic between distribution blocks and out to the WAN and data centre. It should forward quickly and touch traffic as little as possible. | High-end Layer 3 switches |
A useful mental model: the access layer is the front door where users arrive, the distribution layer is where the building’s rules are enforced, and the core is the motorway that carries everything quickly between blocks.
Two-tier and the spine-leaf fabric
The two-tier (collapsed core) model
For small and medium sites, a dedicated core is overkill — it adds devices, links, and cost for traffic volumes that do not need it. The two-tier or collapsed core design merges the core and distribution jobs into a single collapsed-core layer, leaving two layers: access and collapsed core.
- The access layer still connects hosts and enforces port policy.
- The collapsed-core switches aggregate the access switches, route between VLANs, apply policy, and connect to the WAN — performing both the distribution and core roles.
This cuts latency (fewer hops), cost, and management overhead, which is why it dominates small campuses and branch sites. The trade-off is scale: as traffic and the number of blocks grow, a single collapsed layer becomes a bottleneck and a bigger failure domain, and a dedicated core becomes worthwhile again.
Spine-leaf for the data centre
The traditional campus hierarchy was designed for north–south traffic (users reaching servers). A modern data centre is dominated by east–west traffic: servers talking to other servers, virtual machines migrating, storage replicating. For that pattern a different fabric is used: spine-leaf.
A spine-leaf fabric has two device roles: a row of spine switches (the backbone), and a row of leaf switches that connect to the servers and storage.
The rule that defines the design: every leaf connects to every spine, and leaves never connect to each other; spines never connect to each other. Keep it fixed. That rule gives:
- Predictable latency — any server reaches any other server in the same number of hops (leaf → spine → leaf), so a delay-sensitive application behaves the same wherever it is placed.
- Even bandwidth — traffic is spread across all spine links using equal-cost multipath (ECMP), so no link is favoured.
- Easy scaling — add another spine or leaf to grow capacity and ports, without redesigning the fabric.
WAN topologies
A wide area network (WAN) joins sites that are far apart, usually over links leased from a service provider. Three shapes recur, and each is a trade-off between cost and redundancy.
| WAN topology | How it is wired | Strength | Weakness |
|---|---|---|---|
| Hub-and-spoke | Every remote site (spoke) connects only to a central site (hub). | Cheap, simple, easy to manage; one place to apply policy. | The hub is a single point of failure and a bottleneck; spokes cannot talk directly. |
| Full mesh | Every site connects directly to every other site. | Maximum redundancy and best path between any two sites. | Very expensive; links grow as n(n−1)/2. |
| Partial mesh | Some sites are fully interconnected; others connect to one or two. | Balances redundancy against cost; common in real WANs. | Requires careful routing so traffic takes the path you intend. |
Recall from Lesson 12 that the link count for a full mesh of n devices is n(n−1)/2. Four sites need 4×3/2 = 6 links; six sites need 15 links. That rapid growth is exactly why full mesh is reserved for high-value core links and partial mesh is used everywhere else.
Internet connection topologies
How an organisation attaches to the Internet is described with homed terminology. The definitions below follow the widely taught Cisco scheme; note that some vendors use the words slightly differently, so read the answer options carefully in any exam question.
| Term | Connections | Survives… |
|---|---|---|
| Single-homed | One link to one ISP. | Nothing beyond a simple link failure — no redundancy at all. |
| Dual-homed | Two or more links to the same ISP. | A link or device failure, but a whole-ISP outage still takes you down. |
| Multihomed | Links to two or more different ISPs. | A single ISP’s outage — traffic can fail over to another provider. |
| Dual-multihomed | Two or more links to each of two or more ISPs. | Both link and ISP failures; the most resilient and the most expensive. |
Two ideas drive the choice. Link redundancy protects against a cable or provider device failing; ISP diversity protects against an entire provider failing. Only multihoming across different providers gives you the second. Achieving it usually requires the organisation to own an autonomous system number and run BGP to advertise its address block to both providers — a topic beyond CCNA but worth recognising.
The SOHO architecture, and on-premises vs cloud
A small office / home office (SOHO) does not stack separate devices the way an enterprise does. Instead, one inexpensive box — often called a residential gateway or integrated services router — performs every role at once.
| Role | What it does |
|---|---|
| Router | Forwards traffic between the home LAN and the ISP. |
| Switch | Provides a few wired Ethernet LAN ports. |
| Wireless access point | Lets Wi-Fi hosts join the LAN. |
| Firewall and NAT | Blocks unwanted inbound traffic and lets many private hosts share one public address (Week 6 covers NAT). |
| DHCP server | Automatically hands out addresses, masks, and a default gateway to hosts. |
On the outside, the SOHO connects to the ISP through whatever access technology is available: cable, DSL, fibre to the home, fixed wireless, satellite, or cellular. The whole design favours low cost and zero administration over the redundancy and policy control of an enterprise campus.
On-premises vs cloud
Where the servers and services live is itself an architectural choice.
| Aspect | On-premises | Cloud |
|---|---|---|
| Who owns the hardware | The organisation buys, houses, powers, and maintains it. | A provider owns and operates it; the customer consumes it over the network. |
| Cost model | Large up-front capital expense (CapEx). | Ongoing operating expense (OpEx), pay for what you use. |
| Scaling | Slow and costly — buy and install capacity. | Fast and elastic — add or release resources on demand. |
| Control and responsibility | Full control; the organisation owns all security and maintenance. | Shared responsibility; the provider secures the platform, the customer secures what it puts on it. |
Cloud services are usually grouped into three models: IaaS (infrastructure as a service — virtual machines, storage, networks), PaaS (platform as a service — a managed environment to deploy applications), and SaaS (software as a service — ready-made applications such as webmail). Many organisations run a hybrid design, keeping sensitive or legacy systems on-premises while moving elastic workloads to the cloud.
Review
Check what you have learned with these review questions.
Sources & references
- Cisco. CCNA (200-301) Exam Topics, Domain 1.0 Network Fundamentals. cisco.com.
- Cisco. Campus Network Design — Hierarchical Design Model, Cisco Validated Designs. cisco.com.
- NIST. SP 800-145: The NIST Definition of Cloud Computing (2011). csrc.nist.gov.
- IANA. Autonomous System (AS) Numbers registry. iana.org.
Target exams: CCNA 200-301 v1.1 · CyberOps Associate 200-201. Last content verification: 2026-10-11.