Gravity of Ecosystems: Pollinating the Modern AI Orchestrator
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Claims that shifting core contributors into Ray, Slurm, PyTorch over a year produces concrete Kubernetes changes: implemented Dynamic Resource Allocation and a Workload API extension that adds resource-slice, device-class and request-template objects, ties scheduler and cluster-autoscaler to accelerator topology, and eliminates “double scheduling.” Teams run pilot integrations against Ray/Slurm/PyTorch clusters, validate scheduler-aware GPU/TPU placement, and report reduced idle accelerators and simpler notebook-to-production workflows.
AI-written summary · about
chapters (8)
- 0:00Conference start
- 3:49AI and Kubernetes intro
- 8:57Day-2 operations in Kubernetes
- 15:13Cluster autoscaler and load planning
- 22:07Dynamic resource allocation in Kubernetes
- 28:26Kubernetes future and infrastructure risks
- 30:27Kubernetes availability and observability
- 33:02Community contribution and closing
related talks
09mKeynote: Kubernetes for GenAI: The New Operating System of the AI-First World
27mHarnessing Kubernetes for Scalable AI/ML Workloads
27mEvolving and Scaling Ray for the Next Generation of AI Workloads
28mScaling Multimodal Data and Reinforcement Learning with Ray
25mExtending Kubernetes for AI | Lessons Learned From Platform Engineering
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
25mHow We Power the Largest AI Deployments on the Planet: Running Vir...
29mScaling AI workloads with Ray & Airflow