Better Sharing of AI Accelerators in OpenStack with Blazar Reservations

Learn how to optimize GPU resource sharing in OpenStack environments through a 47-minute conference talk from OpenInfra Day France 2024. Explore the challenges of managing GPU devices for AI and machine learning workloads in private clouds, where limited availability and high costs create resource contention issues. Discover how the OpenStack Blazar project enables advanced resource reservation capabilities, with a focus on StackHPC's developments for AI-optimized flavors, GPU accelerator allocation, and CPU pinning enhancements. Presented by Pierre Riteau from StackHPC, delve into practical solutions for improving resource utilization efficiency and implementing better sharing mechanisms for AI infrastructure components in OpenStack deployments.