Let There Be Topology-Awareness in Kube-Scheduler!
summary
Explains hardware topology and Kubernetes Topology Manager, identifies gaps where the kube-scheduler does not use topology information, and surveys existing out-of-tree solutions and workarounds (custom schedulers, scheduling extensions, node selectors, manual resource assignment). Describes the problems caused by scheduler ignorance of NUMA and heterogeneous nodes for performance-sensitive workloads (5G, Telco, Edge, IoT, AI/ML) and outlines plans to enhance the native kube-scheduler to perform topology-aware NUMA-aligned placement.
AI-written summary · about
related talks
18mLet There Be Topology-Awareness in Kube-Scheduler!
24mAccelerate Your AI/ML Workloads With Topology-Aware Scheduling in Kueue
47mNHD - A Topology-Aware Scheduler for K8s for Low-Latency & HPC Applications
10mLightning Talk: Mind the Topology: Smarter Scheduling for AI Workloads on Kubernetes
26mResource Orchestration of HPC on Kubernetes: Where We Are Now and the Journey Ahead!
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
36mNetwork-aware Scheduling in Kubernetes
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?