Building Source Routed AI Backend Network with SRv6

This 16-minute conference talk by Guohan Lu, Partner Software Engineer at Microsoft, explores an innovative approach to building AI backend networks using SRv6 (Segment Routing over IPv6). Learn how this alternative to traditional BGP-based architectures addresses the demanding requirements of AI infrastructure, including higher bandwidth, lower latency, and improved fault tolerance. Discover the design rationale behind implementing a completely source-routed network that gives end hosts ultimate control over network path selection. The presentation covers the high-level architecture and initial implementation details of this cutting-edge networking approach for AI systems presented at the Open Compute Project.