Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore Meta's multi-generational journey in developing AI data center networking architectures through this comprehensive keynote presentation. Learn how Meta has evolved its networking infrastructure to support increasingly demanding machine learning workloads, with insights from Omar Baldonado who leads the development and operation of Meta's global data center networks supporting AI models and the Meta family of apps. Discover the interconnected technology layers including PyTorch and AI frameworks, xPU characteristics and traffic patterns, NIC selection and capabilities, and their network implications. Understand how decisions cascade through the technology stack, from framework behavior influencing hardware selection to accelerator characteristics driving network topology choices and NIC capabilities enabling operational approaches. Gain practical insights into network automation at scale, infrastructure density challenges including power, cooling, and space considerations, telemetry approaches for AI workload visibility, and operational strategies for managing rapid technology transitions while maintaining production stability. Examine architectural decisions, false starts, and breakthrough solutions that emerged from deploying and operating multiple generations of AI clusters in production, including the development of some of the world's largest AI clusters with gigawatt-scale clusters on the horizon. Learn about Meta's contributions to open-source networking through libraries like TorchComms for PyTorch and FBOSS for switches, and their involvement in communities like the Open Compute Project.
