UB-Mesh proposes a hierarchical full-mesh network for AI training clusters
bookwormengr · x · 2026-07-22
This reply points to a paper on UB-Mesh, a hierarchical full-mesh datacenter network for AI training.
The core idea is that every NPU can also act as a router, so traffic can take multiple paths instead of relying on a single non-blocking flat fabric. The attached diagram contrasts shortest-path-first routing with all-path routing to show why the design matters for large training systems.
More from Infra
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11