Kubernetes Dynamic Resource Allocation (DRA)

In this session we will deep dive into Dynamic Resource Allocation, a novel feature that's been in the trenches for quite a while now and that's going to change how Kubernetes can be used for AI Workloads, i.e., running GPUs as worker nkdes, distributing traffic, scheduling and doing all the good things that Kubernetes provides for CPU based workloads.

GPU based workloads are very different compared to CPUs, so they need to be taken care of differently. Kubernetes 1.34 is going to being a few major features of DRA Generally Available. We will try to cover as much of ground as possible during this session.

Please try to bring your laptops so that we can together work on some exercises.