Learn Backend Development Part-Time, Online
Gain a Splash of New Skills - Coursera+ Annual Nearly 45% Off
Overview
Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Learn about Congestion Signaling (CSIG), a cutting-edge network technology designed to address the unique challenges of modern AI workloads in this 16-minute conference talk from the Open Compute Project. Discover how traditional network congestion signals fail to capture the detailed fabric bottleneck conditions needed for AI workloads that send millisecond-duration traffic bursts, and explore how CSIG provides a revolutionary solution through always-on, line-rate telemetry in application context. Understand the technical implementation of CSIG, which offers sub-millisecond-granularity switch traffic measurements and fine-grained, per-packet feedback for congestion control that captures both the congestion state and location of bottlenecks along network paths. Examine real-world deployment results from Google's implementation, including demonstrated performance improvements to congestion control and load balancing systems. Gain insights into how CSIG telemetry uniquely enables understanding of real-world workload behavior and learn about the standardization efforts currently underway at UEC with support from multiple switch vendors.
Syllabus
CSIG Congestion Signaling in the AI Era
Taught by
Open Compute Project