Slinky Expanded: Slurm, Kubernetes, and DRA - Praveen Krishna, Google & Marlow Warnicke, SchedMD LLC

This talk features Praveen Krishna from Google and Marlow Warnicke from SchedMD LLC as they discuss enhancing performance and efficiency for training and multi-node inference jobs through advanced resource scheduling. The session focuses on the slurm-bridge scheduler from the SlinkyProject, which integrates with Kubernetes to optimize workload management. By utilizing Dynamic Resource Allocation (DRA) within Kubernetes, the speakers demonstrate how to effectively use detailed resource information for improved cluster utilization and reduced costs. Attendees will gain insights into the collaboration of these technologies and how they can lead to greater operational efficiency in cloud native environments.