Enabling HPC and ML Workloads with Latest Kubernetes Job Features

Explore the latest Kubernetes Job API features for running distributed Batch, AI, and HPC workloads at scale in this conference talk. Learn how Indexed Jobs simplify parallel workloads requiring pod-to-pod communication, with examples from DeepMind's distributed machine learning applications. Discover the Flux Operator's ability to orchestrate HPC workloads by creating a "Mini Cluster" within Kubernetes. Understand how Pod Failure Policy can maintain job execution despite pod disruptions while optimizing costs. Gain insights from real-world experiences at DeepMind and Lawrence Livermore National Laboratory to enhance your ability to manage complex computational workloads in Kubernetes environments.

Enabling HPC & ML Workloads with the Latest Kubernetes Job Features- Michał Woźniak & Vanessa Sochat