Kubernetes DRA Architecture: Scheduling, Status, and Topology at Scale
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Kubernetes Dynamic Resource Allocation (DRA) in v1.36 is presented as a unified architecture covering entry points, resource modeling, scheduler allocation and binding, kubelet preparation, status and health, and access boundaries. It focuses on topology-aware scheduling for supernode AI/HPC systems (NUMA, PCIe, NVLink, fabric) and includes a case study of IMEX daemon/channel assignment that moved from distributed-conflict updates to centralized updates and then to distributed updates on a topology-aligned state object, cutting assignment latency from minutes to seconds at thousand-device scale.
AI-written summary · about
related talks
27mUncharted Waters: Dynamic Resource Allocation for Networking
29mWhat Can I Get You? An Introduction to Dynamic Resource Allocation
50mDynamic Resource Allocation - the path towards GA
42mGuide Us on Next Steps in Kubernetes Device Manage...
30mKubernetes WG Device Management - GPUs, TPUs, NICs and More With DRA
30mDRA in KubeVirt: Overcoming Challenges & Implementing Changes
24mFit-to-Serve: How a New DRA Capability for Dynamic Device...
32mLeveraging DRA for Conquering ML Workload Challenges on...