Why Fat-Tree, Dragonfly, and Torus Topologies Matter for HPC Networks

The article analyzes three major high‑performance‑computing network topologies—Fat‑Tree, Dragonfly, and Torus—detailing their design principles, scalability formulas, routing strategies, advantages, and limitations to help architects choose the most suitable architecture for large‑scale GPU clusters.

DragonflyFat-TreeGPU clusters

0 likes · 13 min read

Why Fat-Tree, Dragonfly, and Torus Topologies Matter for HPC Networks

Architects' Tech Alliance

Mar 25, 2024 · Industry Insights

Why Fat-Tree, Dragonfly, and Torus Topologies Matter in HPC Networks

The article examines the challenges of ultra‑large‑scale HPC networking, compares traditional CLOS with Fat‑Tree, Dragonfly, and Torus topologies, explains their bandwidth and latency characteristics, presents scalability formulas, and evaluates routing algorithms and practical trade‑offs for each design.

Data CenterDragonflyHigh‑performance computing

0 likes · 14 min read

Why Fat-Tree, Dragonfly, and Torus Topologies Matter in HPC Networks

Architects' Tech Alliance

Mar 20, 2023 · Fundamentals

Dragonfly Network Topology and Routing Algorithms for High‑Performance Data Centers

The article explains the Dragonfly network topology, its hierarchical structure, key parameters, routing algorithms (Minimal, Non‑Minimal, UGAL variants) and deadlock avoidance techniques, highlighting how modern data‑center networks address latency bottlenecks in high‑performance computing environments.

Data CenterDragonflyHigh‑Performance Computing

0 likes · 9 min read

Dragonfly Network Topology and Routing Algorithms for High‑Performance Data Centers