Kubernetes Podcast from Google HPC Workload Scheduling, with Ricardo Rocha
Jul 9, 2025
Ricardo Rocha leads the Platform Infrastructure team at CERN, specializing in cloud-native deployments and machine learning. He discusses CERN's shift to efficient computing solutions and the integration of Kubernetes for scientific workloads. The conversation highlights ongoing collaborative efforts to enhance workload scheduling and resource management. Ricardo also delves into the evolution of Kubernetes, its advantages for high-performance computing, and the critical role of user feedback in shaping its development and features.
AI Snips
Chapters
Transcript
Episode notes
Flying Planes and Planning Kubernetes
- Ricardo Rocha relates flying airplanes to cloud native planning, emphasizing the need for preparation.
- Both disciplines benefit greatly from advanced planning to avoid repeated mistakes and ensure smooth operations.
Kubernetes' Original Limits for Science
- Scientific workloads rely on batch jobs, requiring advanced scheduling unlike typical service workloads.
- Features like queues, preemption, and priorities needed for scientific computing are missing from original Kubernetes.
Early Adoption of Kueue Scheduler
- Engage early with emerging projects like Kueue to meet scientific workload needs in Kubernetes.
- Collaborate with community and organizations to shape schedulers that support batch and HPC workloads.