Sailing Ray Workloads with KubeRay and Kueue in Kubernetes

Explore how to manage Ray workloads in Kubernetes using KubeRay and Kueue in this informative conference talk. Learn about the growing compute demands in machine learning and how Ray, a unified computing framework, enables ML engineers to scale workloads without complex infrastructure. Discover the benefits of using Kubernetes with KubeRay for managing diverse workloads, and gain insights from ByteDance's experience of submitting thousands of jobs daily to Ray clusters. Understand the challenges of managing concurrent Ray jobs, including job starvation and resource allocation, and how Kueue, a Kubernetes native job queueing system, addresses these issues with features like resource management, multi-tenant support, and fair-sharing.

Sailing Ray Workloads with KubeRay and Kueue in Kubernetes - Jason Hu, Volcano Engine & Kante Yin