Announcing Kubernetes Dynamic Resource Allocation for Elastic Fabric Adapter

Amazon EKS EFA DRA Driver Support

Amazon Elastic Kubernetes Service (Amazon EKS) now supports Dynamic Resource Allocation (DRA) for Elastic Fabric Adapter (EFA), enhancing high-performance inter-node communication and RDMA for AI, ML, and HPC workloads. The EFA DRA driver, based on the DRANET project, enables EFA interface sharing and topology-aware allocation for Kubernetes workloads.

With the EFA DRA driver, you can allocate EFA interfaces and accelerator devices sharing the same PCIe root or device group, ensuring inter-node traffic flows through the closest network interface to each GPU, AWS Trainium, or AWS Inferentia device on the node. It also supports EFA interface sharing across workloads on the same node to maximize EFA interface utilization.

What to do

• Use the EFA DRA driver for new deployments on Amazon EKS clusters running Kubernetes version 1.34 or later.
• The EFA device plugin remains supported for use with Karpenter and Amazon EKS Auto Mode.

Source: AWS release notes

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.