Skip to content

6–12x Route Programming Acceleration in SONiC: A Collaborative Profiling and Optimization Journey#

OCP Global Summit 2026 digest Program Core pick 12–15 October 2026

Abstract Session recording Materials

Deepak Singhal (Microsoft), Amol Rawal (Nokia)

  • Why: Contributor-led SONiC internals with hard numbers (2K–4K routes/sec baseline → 6–12x), a reproducible benchmarking methodology, and upstreamed code — the archetype of a talk that reveals where an open-source project is actually heading.
  • What: A joint Microsoft–Nokia effort that built a Snappi-based benchmarking framework, profiled existing optimization proposals and their combinations across multiple hardware platforms, then contributed fresh optimizations: replacing Redis-based IPC with direct ZMQ communication, offloading synchronous recording from the hot path, and integrating tcmalloc into FRR daemons (from abstract only).
  • How: Measurement-first — the framework identifies which optimizations actually matter and how they interact, rather than merging proposals blindly; results were driven through SONiC Routing Working Group consensus with published HLDs across multiple sonic-net repositories.
  • Where applicable: SONiC on any supported platform (multi-platform profiling included); most relevant to fabrics pushing 1M+ route scale, e.g. large AI clusters with heavy ECMP.