Which Network Topology?

Recently Iíve come across some†interesting terms for variants of common†network topologies, so I decided†I’d try to list†as many of them†as I can for reference. Please suggest others to add.


In the bus topology, devices connect directly to a common backbone. For instance, the early days of Ethernet saw all devices connected directly to a single piece of coax cable using BNC T-connectors. Can anyone think of any instances of bus topologies in modern networks?


SDH / SONET technologies spring to mind when I think of†the ring topology, or maybe FDDI and token ring.

Thereís also the dual-ring topology variant, which provides a second ring for redundancy.

A triangle can be considered one extreme of a ring topology, which is a recommended building block for campus and data centre LAN designs.


The linear topology is just a number of nodes in series. It is also known†by names incuding†line, string of pearls and daisy chain. Unfortunately I’ve seen this topology a few too many times in networks†that have been allowed to sprawl uncontrolled.


The star topology is where all nodes connect back to a central node. It’s common†in the LAN environment, for instance where PCs connect back to an access switch.

There’s also the†snowflake topology, which†is a ‘star of stars’.


The tree topology provides scalability. Itís the basis of most enterprise and service provider†backbone networks, allowing large numbers of devices to be connected together in a hierarchical topology.

Recently Iíve come across the term Fat Tree being used†in terms of data centre and campus designs. This is where the links become ëfatterí, ie provide more bandwidth as one moves up towards the root of the tree. For instance, a PC could be connected to an access switch at 100Mb, which then has a 1Gb uplink to a distribution router, which in turn has a 10Gb uplink to a core router.

Full Mesh

The full mesh topology is where all nodes are connected directly to all other nodes. Although this provides the most direct path between endpoints, in reality it doesnít scale for anything but the simplest architectures.

If†N is the number of nodes, then N(N-1)/2 links are required to provide a full mesh. The diagram shows 8 nodes and 28 links to create a full mesh,†while a topology†with 100 nodes would require†4,950 links!

Partial Mesh

The partial mesh is simply a subset of the full mesh links. This topology is most commonly associated with the WAN when†connecting branch offices to one or more hub sites.

The Single Hub and Spoke is arguably the extreme of the partial mesh topology (it would also†become a†star topology), where all nodes connect only to a single hub site. All communication between spoke sites must be via the hub. The main problem with this topology is lack of resilience if the hub site goes down.

Dual Hub†and Spoke topology is the most common in an Enterprise WAN environment. Each spoke (branch site) connects to two hubs (central sites), providing resilience if one hub goes down.

Dual Plane Topology

This topology is†the duplication of two other network topologies, generally to improve resilience by eliminating fate sharing between them. For instance, an Enterprise customer could buy WAN services from two service providers and†connect their sites to both of them†for resilience. Some service providers even build out duplicate networks themselves,†sometimes using different vendor equipment for each.


Hybrid is†a combination of topologies that do not exhibit the characteristics of any standard topology.

  • http://twitter.com/ziyadb_ Ziyad Basheer

    Hey Tony, heard back about the CCDE yet?

    Ziyad B.

    • Tony Brown

      Hi Ziyad,

      Not yet, they seem to try and string it out for as long as possible. It normally takes about 8 or 9 weeks, so hopefully I should hear back shortly.

      Enjoy the trip to Dubai!

  • http://ignore-your.tv/ James Cape

    Don’t all internetworks (i.e. L3) of sufficient size behave as an active bus?

  • Den

    Isn’t a DOCSIS network a merge between a star and a bus?

    • Tony Brown

      Oh yes, great example, thanks!

  • http://unroutable.blogspot.com jswan

    There’s also “multiplanar”, which is like the dual-plane topology, except that the two planes are connected at key points. Best visualized in 3-D.

    • Tony Brown

      I like this one, I’ll add it if I can work out how to draw it! :)

      Any examples of where this is used?

      • http://unroutable.blogspot.com jswan

        John Cavanaugh from Cisco talks about it a lot in his Networkers presentation on the evolution of high availability network design.

  • Brannen

    Sneaker net. 2 or more nodes, not connected, where data is manually transferred from one to the other. Any bearer can conduct the data, including carrier pigeon, as specified in RFC1149. http://www.ietf.org/rfc/rfc1149.txt

    • Tony Brown

      I forgot the most fundamental of all, thanks!

  • http://layer3.wordpress.com Chris Church

    Good post. I understand the concept of Fat Tree but it seems like a stretch to consider it a seperate topology.

    • Tony Brown

      Thanks Chris! I didn’t mean to imply Fat Tree is a separate topology, I just wanted to describe it.

  • Ryan Malayter

    Any thoughts about revisiting this with Clos, folded-Clos, butterfly, and flattened butterfly topologies now that TRILL/SPB is coming? (In reality such topologies can be had with layer 3 switches today, and I think they are alreeady used in the HPC/big cloud space).

    Tree networks can’t support the any-to-any communication model well. Even fat-tree networks are un-build-able at a certain scale. Even if 16x40Gbps trunks near the root were feasible economically, that would still only support 640 gigabit-connected servers with full bistection bandwidth.

  • Er Rajivkumarjohn

    There is a Mesh topology using 8 node send the data from 2 to  6 what happen when 5 stope working. What will happen then?