Spine and leaf architecture - connections to spine switches

In a network employing spine and leaf architecture, spine switches don't need to be connected to each other. Actually, by definition, they should not be connected to each other. Spine switches should only be directly connected to leaf switches. Indeed, spine switches should not even have any uplinks either!

Spine switches should only be connected to leaf switches or:

1. When you have a multi-pod environment, where you want to connect two or more datacenters to each other, where both datacenters are part of the same fabric. In this case, we use what is known as an interpod switch, that can connect spine switches from two different "pods". But these pods are managed by the same APIC.
2. In a multi-site environment, which is similar to a multi-pod environment. The only difference is in this case, you have two separate fabrics controlled by two different APICs.
3. A spine switch can also connect to a remote leaf. A remote leaf enables you to extend your fabric into a satellite site where you don't want spine switches.

If you want to learn more about how to connect leaf switches, check out this note:

Spine and leaf architecture - connections to leaf switches

One of the reasons why we connect spine switches in a certain way is to reduce latency:

Spine and leaf architecture - reducing latency