When we think about building networks, it’s important to understand and be able to choose between different physical topologies. In this post, i’ll go over some of the common ones used in today’s networks and i’ll identify the advantages and disadvantages for each one them.
The physical topology of the network is the foundation on top of which you will build your design. Let’s take a look at some of the most common ones and in which part of the network they are found.
Hub and Spoke
Hub and spoke is a very popular WAN centralized model that is low cost since it only requires each spoke to have one connection back to the hub. One of the most important design considerations for this topology is to make sure you summarize or filter prefixes properly to avoid spokes going through other spoke’s to reach a remote site. The hub and spoke is the easiest topology to implement due to it’s simplicity of adding additional nodes and it’s low number of neighbor counts at each spoke. The biggest disadvantages for this model is that it offers no redundancy in case of link failure and that the scalability of the design will rely heavily on the hub’s capabilities.
The ring is the cheapest closed loop topology available as every device in the topology has only 2 neighbors to each other. It is most widely seen in core or WAN modules but is also very popular as a last mile service provider topology. Two main disadvantages to this model:
- There will be convergence issues in the PQ space causing micro-loops. This results in slow convergence by default with most routing protocols. REP or ERP can be implemented for Layer 2 fast convergence as these protocols were designed for ring topologies.
- It is prone to congestion when elephant flows are present. Traffic Engineering is suggested in most cases to avoid this issue.
The partially meshed topology is often the result of organic growth in the WAN or core network layer. To understand how this type of topology converges better, we can break it into a series of hub and spoke and rings. This type of network will, in most cases, offer a balance of redundancy, scalability and cost.
Fully meshed is one of the most costly WAN or core topology designs in terms of CAPEX but can be required if high availability and low latency is needed. One of the main disadvantage of this topology is the added control-plane complexity and scalability problems that are inherently present when you have high number of nodes fully interconnected.
The hierarchical topology combines multiple hub and spoke networks together. This type of topology is often used in aggregation modules of data-center and campus designs or large WAN’s to scale the network through summarization and topology hiding.
The disjoint core or also called the disjoint parallel plane is a topology often created to separate data-planes. The two inner rings of this model are designed in such a manner that there’s no way to forward traffic from one to another. Often this type of architecture will be used when high availability is required and it will also be paired with separate vendor and control-plane protocols. For example, the outer ring could run Juniper devices + IS-IS as a routing protocol and the inner ring could run Cisco devices + OSPF.
Spine/Leaf or Clos
The last model that we’ll look at in this post is called the Spine and Leaf or clos topology. It was formalized by Charles Clos in 1952 and most recently adopted by most hyperscale data-centers. The key advantage of this type of network is it allows for very low over-subscription ratio’s in a non-looped topology by making sure that the leafs (lower devices) and spines (upper devices) are not interconnected between each other. This topology promotes east-west traffic flows rather than the north-south like the older hierarchical data-center designs. Since every leaf is connected to every spine, ECMP is used and this results to little to no effect when a failure occurs in terms of convergence times or latency.
Here’s a comparison chart for each topology comparing cost, manageability, redundancy, scalability, low latency after a failure and optimal path between each nodes. Cost refers to the CAPEX for implementation and manageability refers to the OPEX (complexity of maintenance and troubleshooting). Scalability on the other hand represents how well routing and switching protocols can handle an increasing number of nodes while keeping convergence times and impact to the control-plane low.
In conclusion, it is up to the network designer to choose the right type of topology based on the business requirements and to understand the technical trade-off for each one of them. Please keep in mind the comparison chart provided is only a guideline and that the final architecture decision should take into account multiple other factors driven by the business needs.