The challenge
sciCORE is the University of Basel's centre for scientific computing. Workloads like that put sustained, heavy demand on the network, and a network that has grown organically over the years tends to answer that demand with patches, not structure.
Performance bought with complexity becomes a liability. Every exception, every one-off link and every undocumented path is something that has to be understood again the next time the network is touched. The goal was a design that delivers the performance, and stays manageable while doing it.
The approach
We approached it as a design exercise, not a patching exercise, and proved the design before it ever touched production.
-
Design for the structure, not the symptoms
Rather than extend the grown topology, we designed a clean, scalable structure: a spine-leaf fabric where capacity is added by adding leaves, and where the path between any two points is predictable.
-
Validate it in the lab
The design was built and tested in a network lab first: a 100G spine-leaf fabric with MLAG, modelled end to end so its behaviour was known before a single production device was reconfigured.
