OCP 2026 - OpenAI: Clustermapper: Deterministic path diagnosis for AI backend networks using SRv6
At the OCP Global Summit 2026 (opencompute.org), Ryder Lewis and Guohan Lu of OpenAI presented “Clustermapper: Deterministic path diagnosis for AI backend networks using SRv6.”
Once an AI backend fabric uses SRv6 for strict source routing, operators can also use those same path identifiers for diagnosis. Clustermapper maps and tests the deterministic paths that training traffic actually takes, rather than inferring them from ECMP hashes.
The session is a call to action across the ecosystem: switch and ASIC vendors and SONiC contributors to extend and qualify SRv6 steering—including longer paths with SRH and a supported way to select individual links when a device has multiple adjacencies; and operators to make deliberate path testing part of fabric acceptance, covering workload links, identifying gaps, and verifying payload integrity.