One Cluster, End-to-End
summary
Demonstrates collapsing data prep, distributed training, and serving onto one Amazon EKS cluster using SageMaker HyperPod, running heterogeneous CPU/GPU Ray clusters with queue-based admission and gang and elastic admission so large jobs either start whole or resize without losing Ray state. Shows in-cluster fast checkpoint tiering that syncs to durable storage, node-failure detection and automated reboot/replacement to avoid pipeline restarts, namespace multi-tenancy, VS Code/Jupyter dev environments, and built-in observability to raise GPU utilization.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:51Queuing for resource sharing
- 6:05Managed developer environments
- 8:13Multi-tenant interface
- 8:34Resilient infrastructure
- 11:22Ray on SageMaker HyperPod
related talks
14mBuilding a high-performance analytics ecosystem on Amazon EKS
47mCloud Native and Sustainable, Reproducible Scientific Computing
27mGen-AI at Scale: Simplifying Orchestration of Healthcare Applications Across Multi-Cluster Environme
46mMachine Learning with Kubernetes
28mGPU sharing done right: Secrets, security, and scaling with Vault and vCluster
45mAccelerate Model Training with an Easy to Use High-Performance...
29mTowards Sustainable Multi-Cluster Management in Real-World Academic Infrastructure
36mNavigate the future of AI: Deploy generative models on Amazon EKS