Fat-Tree topologies are hierarchical network architectures commonly used in data centers and HPC environments to provide scalable, high-bandwidth connectivity between large numbers of endpoints. They are designed to deliver predictable performance while supporting simultaneous communication across the network.

The term "fat tree" refers to the increased bandwidth capacity in the upper levels of the network hierarchy, which appear "fatter" than the lower levels when represented diagrammatically. Unlike a traditional tree topology, where available bandwidth decreases as traffic moves toward higher levels of the hierarchy, a Fat-Tree topology maintains consistent bandwidth throughout the network. This is achieved by increasing the number of links, the link capacity, or both at each higher tier of the topology. As a result, the aggregate bandwidth available between groups of endpoints remains balanced across all levels of the network.

By reducing oversubscription and providing multiple paths between endpoints, Fat-Tree topologies can support high-throughput, low-contention communication patterns. These characteristics make them well suited for demanding workloads such as HPC, AI training, large-scale data analytics, and cloud computing, where many systems must exchange large volumes of data concurrently. This additional capacity helps ensure that network performance remains consistent as traffic flows through the fabric, enabling scalable and efficient communication across large deployments.