Addressing Endpoint-Induced Congestion in Scale Out Accelerator Domain
OpenFabrics Alliance via YouTube
Lead AI Strategy with UCSB's Agentic AI Program — Microsoft Certified
Stuck in Tutorial Hell? Learn Backend Dev the Right Way
Overview
Google, IBM & Meta Certificates – 40% Off
One plan covers every Professional Certificate on Coursera.
Unlock All Certificates
Learn about innovative RDMA mechanisms for handling endpoint-induced congestion in accelerator domains through this technical presentation from Intel researchers. Explore a new UDP/IP/Ethernet-based approach designed specifically for lossy networks without PFC-like mechanisms, targeting medium-scale domains of hundreds of nodes. Discover how this FPGA-implemented solution achieves 200Gbps reliable bandwidth while using minimal resources, and examine novel heuristics that modify ACK mechanisms to improve congestion handling. Understand the challenges of accelerator scaling, including bandwidth requirements, latency constraints, and endpoint congestion issues, while gaining insights into practical solutions that enhance goodput and reduce packet loss compared to conventional methods.
Syllabus
Addressing Endpoint induced Congestion in a Scale Out Accelerator Domain
Taught by
OpenFabrics Alliance