I was lucky to see a tweet by Dónal today that made me read more about CLOS networks.
Network failure domains. Open and closed networks have different properties. Fat trees = top down comms optimised. Check out CLOS.
— Dónal (@irldexter) July 15, 2018
The CLOS network removes the bottlenecks inherent in the main nodes of a classical hierarchical system.
Instead, it expands the amount of communication that occurs between leaves and the spine.
If there isn’t any oversubscription taking place between the lower-tier switches and their uplinks, then a non-blocking architecture can be achieved.
CLOS architecture allows each leaf to be only two hops away from another one. And if any spine node goes down, communication is still working. Bandwidth is increased by growing the spine and growing the leaves.**
Why is this important to me? It’s super relevant to something I’m working on with respect to scaling design in an all-remote team at Automattic. —JM