Share With Care - Efficient Device Sharing With Guaranteed Resources Using DRA

Explore how Kubernetes can efficiently share network hardware and other devices across multiple Pods using the new Device Resource Allocation (DRA) framework in this 32-minute conference talk. Learn to overcome the traditional challenges of dividing network cards, creating virtual interfaces, limiting bandwidth, and distributing resources to different Pods through on-demand provisioning based on resource requests. Discover how this functionality extends beyond networking to work with any device type including NICs, GPUs, and DPUs, enabling physical devices to be independently shared among multiple Pods while maintaining resource guarantees. Examine the technical implementation details of how DRA performs sub-device level resource allocation, effectively extending Kubernetes multi-tenancy capabilities to individual device components. Gain insights into real-world use cases where this approach significantly improves hardware utilization, particularly for expensive specialized equipment, and explore future extensions and possibilities for device sharing in cloud native environments.