Network Topology Generator (ASCII)

Network topology fundamentals

A network topology describes how devices (hosts, switches, routers, firewalls, load balancers) are connected — physically by cables/fibers and logically by the protocols and address spaces that flow over them. Picking the right topology drives fault tolerance, bisection bandwidth, latency, blast radius and cost. A good topology diagram is the first artifact a SRE, network engineer or auditor looks for when triaging an incident or assessing a change.

There are two layers of meaning: the physical topology (where the cables actually go) and the logical topology (how packets flow once protocols and VLANs are configured). A switched Ethernet network with one core switch is physically a star but may be logically a flat broadcast domain, a hub-and-spoke L3 with VLANs, or a multi-tenant overlay.

Classical LAN topologies

• Bus — all hosts on a shared backbone (legacy 10BASE2 Ethernet, single point of failure)
• Star — central switch/hub, each host on its own link (modern Ethernet)
• Ring — each device connected to two neighbors (Token Ring, FDDI, some MetroEthernet)
• Mesh — every device connected to every other; N×(N-1)/2 links full-mesh, expensive but max redundancy
• Tree — hierarchical (core / distribution / access in Cisco's classic three-tier)
• Hybrid — mix of the above, the usual real-world case

Datacenter and WAN topologies

Modern datacenters favor Clos / Spine-Leaf topologies: a leaf layer of top-of-rack switches and a spine layer where every leaf connects to every spine. The result is a two-hop network with predictable east-west bandwidth and ECMP-friendly forwarding — the basis of Fat Tree and dragonfly variants used in HPC and hyperscale clouds. WANs use point-to-point links, hub-and-spoke (very common in SD-WAN), full mesh between regions, or partial mesh between PoPs.

Documentation layers

A complete network diagram set covers all OSI layers: L1 shows physical cabling, rack U positions, patch panels, Cat 5e/6/6a/7 copper and MM/SM fiber runs. L2 shows VLANs, STP/RSTP/MSTP topology, link aggregation. L3 shows subnets, BGP/OSPF/EIGRP areas, routing policy. L4-L7 shows firewall zones, NAT, load balancers and application paths. Mixing layers in one diagram is a common anti-pattern that produces visuals nobody can read.

Tooling

Industry standards: draw.io / diagrams.net (free, ships official Cisco/AWS/Azure/GCP icon stencils), Lucidchart, Microsoft Visio, NetBrain. For simulation: GNS3 and EVE-NG run real Cisco/Juniper/Arista images; Cisco Packet Tracer is the educational baseline. Mermaid and PlantUML render fast from text but lack real network icons. Auto-discovery from a live network is possible with SNMP, LLDP, CDP and NetFlow/sFlow exports — tools like LibreNMS, Observium and SolarWinds NPM build topology maps from this telemetry.

FAQ

Does this generator render real icons or just provide syntax? Generators differ: some output a finished image, others emit a Mermaid/Graphviz/draw.io XML snippet you import. Ours produces a quick visual sketch — for production-grade icon-rich diagrams use draw.io with the Cisco stencil set.

Is it useful for SD-WAN documentation? Yes, as long as you keep separate diagrams per layer. SD-WAN overlays typically need a logical hub-and-spoke or full-mesh diagram on top of an underlay Internet/MPLS diagram — drawing both in one canvas confuses readers.

Can I auto-discover my topology? Yes, but it requires an agent or read-only SNMP/SSH access to the devices. LibreNMS, NetBox plus the napalm driver, or commercial tools like NetBrain crawl LLDP/CDP neighbors and stitch the graph automatically. Cloud topologies use the provider APIs (e.g., AWS VPC Lattice or aws ec2 describe-vpcs).

Star vs mesh — which to choose? Star is cheap and simple but the central switch is a single point of failure. Full mesh has maximum redundancy and the most cables (N(N-1)/2) — only practical for small node counts or critical backbones. Spine-leaf is the modern compromise.