Learn about Meta's journey evolving FBOSS, their network switch management software stack, for next-generation AI infrastructure in this technical conference talk. Explore the unique challenges of handling network traffic in AI fabric, particularly focusing on elephant flows and low entropy issues. Discover how Meta's engineering team addressed both dataplane and control plane challenges, implemented SAI enhancements, and adapted their infrastructure to support AI workloads. Gain insights from Meta's software engineers and Broadcom's senior leadership as they share practical learnings and solutions from deploying one of Meta's largest services across their datacenters.
Overview
Syllabus
5781 Evolving FBOSS for the Next Gen AI Fabric
Taught by
Open Compute Project