Simplicity and scalability are the first and most critical requirements for networking.


Configuration of the switches inside a data center must be simple. General or global configuration such as AAA, SNMP, syslog, NTP, and others should be replicated line by line, independent of the position of the switches. A central management capability to configure all switches at once is an alternative.


Scalability factors include, but are not limited to, the following:

  • Number of racks supported in a fabric. 

  • Amount of bandwidth between any two racks in a data center. 

  • Number of paths from which a leaf switch can select when communicating with another rack.

The total number of ports available across all spine switches and the oversubscription that is acceptable determine the number of racks supported in a fabric. Different racks may host different types of infrastructure, which can result in different bandwidth requirements.

  • Racks with storage systems might attract or source more traffic than other racks. 

  • Compute racks, such as racks hosting hypervisors with workloads or virtual machines, might have different bandwidth requirements than shared edge and compute racks, which provide connectivity to the outside world. 

Link speed and the number of links vary to satisfy different bandwidth demands. You can vary them for each rack without sacrificing other aspects of the leaf-and-spine architecture.

Figure 1. Pod Network Design

The number of links to the spine switches dictates how many paths for traffic from this rack to another rack are available. Because the number of hops between any two racks is consistent, the architecture can utilize equal-cost multipathing (ECMP). Assuming traffic sourced by the servers carries a TCP or UDP header, traffic distribution can occur on a per-flow basis.