solvers.TopologyModel

solvers.TopologyModel()

Models bisection bandwidth for different network topologies (Wall 10).

This model calculates the effective bandwidth available to collective communication operations based on the physical network topology. Different topologies trade cost against bisection bandwidth — the minimum bandwidth across any cut that divides the network in half.

Literature Source: 1. Leiserson (1985), “Fat-Trees: Universal Networks for Hardware-Efficient Supercomputing.” (Fat-tree bisection bandwidth.) 2. Kim et al. (2008), “Technology-Driven, Highly-Scalable Dragonfly Topology.” (Dragonfly topology model.) 3. Dally & Towles (2003), “Principles and Practices of Interconnection Networks.” (Torus and ring analysis.)

Methods

Name Description
solve Solves for effective network bandwidth under a given topology.

solve

solvers.TopologyModel.solve(fabric, topology='fat_tree', num_nodes=64)

Solves for effective network bandwidth under a given topology.

Parameters

Name Type Description Default
fabric NetworkFabric The network fabric specification (link bandwidth, oversubscription). required
topology str Network topology (‘fat_tree’, ‘dragonfly’, ‘torus_3d’, ‘ring’). 'fat_tree'
num_nodes int Number of nodes in the network. 64

Returns

Name Type Description
Dict[str, Any] Effective bandwidth, bisection fraction, and average hops.
Back to top