Campus, WAN, SOHO, cloud, and spine-leaf topologies

A campus network connects buildings or work areas within a single site. It uses a hierarchical design with three main layers: access, distribution, and core. The access layer connects user devices like computers and phones. The distribution layer joins several access areas and separates their failure domains, meaning a problem in one access area usually does not affect others. The core layer connects distributions and provides fast data routing across the campus. Campus networks often use redundant links and fast protocols to restore service rapidly after failures. This redundancy limits the impact of failures to smaller sections, enhancing reliability.

• Campus networks use hierarchical layers: access, distribution, and core.
• Failure domains are limited by design layers, reducing impact from device or link failure.
• Redundancy and fast protocols help restore connectivity quickly after failures.
• Access layer connects user devices; distribution separates failure domains.
• In the lab, campus PCs communicate directly via the switch when connected and properly configured.

A wide area network connects separate locations or branch offices over long distances. It typically uses routers to join smaller networks at each site through point-to-point or multipoint links, often using public or private data carriers. Traffic paths in a WAN flow from a device at one site to a branch router, then possibly through intermediate routers to a remote router and server. Failure domains in a basic WAN depend on the path shown; if a key link or router fails and no alternate is shown or configured, communication between sites is lost. Redundant WAN designs with extra paths can allow for rerouting, but in a single-path WAN like the example, a failure along that path stops traffic.

• WAN connects separate geographic sites using routers and long-distance links.
• Traffic travels through multiple routers between sites.
• Failure of a link or router in the shown path stops communication unless redundancy is built and configured.
• Routing protocols help reroute only when alternate configured paths exist.
• This paper exercise follows only the main depicted path; alternate paths are outside of this example.

SOHO means small office/home office networks with fewer users and simpler devices than enterprise networks. Typically, a single SOHO router connects multiple devices in a small area and handles routing for different local networks. In these networks, traffic between a PC and a server passes through the SOHO router, which also provides Internet access. If the SOHO router stops working, devices on different local networks shown in the diagram lose communication with each other. This centralized, simple layout offers ease of setup but creates a single point of failure at the router.

• SOHO networks serve a small number of users with simple, central routers.
• Devices on different local networks communicate through one router acting as a gateway.
• Failure of the single SOHO router disrupts communication between the shown devices.
• The traffic path in this example relies exclusively on the router shown.
• SOHO networks prioritize simplicity, but this design increases vulnerability to single-device failure.

Cloud computing delivers services like servers and storage over a network, usually the Internet. The provider manages the infrastructure, including hidden details of redundancy and network isolation. Each device uses a gateway to reach the cloud server; what happens inside the cloud is not shown to users. In these diagrams, failure domains cover the user device, gateway, and path to the visible cloud server. Internal failover or redundancy mechanisms that a provider may use are not shown or tested in these paper exercises. Users cannot observe how a failure within the provider infrastructure is handled or if traffic reroutes.

• Cloud services use provider-managed infrastructure that is abstracted from users.
• Traffic path in this model is from user device to gateway to cloud server.
• Users have no control or visibility into the provider's internal redundancy.
• Paper exercises do not observe or test cloud failover or internal logic.
• Failure is visible only in the drawn part of the network—provider response is outside this model.

Leaf switches connect endpoint devices. Spine switches connect the leaves to one another. In this drawing, each leaf connects to both spines. Equal-cost multipath (ECMP) allows a routing device to use multiple eligible routes to the same destination when those routes have the same routing cost. Cost is a route-selection value, not a money price. Traffic is commonly assigned by flow, meaning related packets such as one download. ECMP does not duplicate each packet across every spine or guarantee an even split of traffic.

Lab boundary: This spine-leaf comparison is a paper exercise. It does not measure routed ECMP in GNS3, load distribution, failure-detection time or routing convergence. Convergence means updating forwarding choices after the network changes. GNS3's built-in Ethernet switch provides switching, not the routed ECMP implementation needed for this experiment. A routed test needs capable, properly licensed devices and a separate configuration. The live campus ping exercise tests only its local path.

• Equal-cost routes are eligible routes to the same destination with the same routing cost; extra cables alone do not enable ECMP.
• A flow commonly uses one selected path. Multiple eligible paths do not guarantee equal traffic shares or copy every packet to every spine.
• Both drawn paths end at Spine Server. Losing its only leaf removes this server's paths, not necessarily those of every server in a real fabric.
• The illustrated inter-leaf path crosses one spine and two fabric links. Same-leaf traffic need not cross a spine; this is not a universal IP hop-count rule.
• This paper comparison does not measure routed ECMP in GNS3. Load balancing, routing convergence and recovery timing remain untested.