Sign up to save your podcasts
Or
Share Making Kubernetes Work for AI and Batch Workloads with Kevin Hannon
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Making Kubernetes Work for AI and Batch Workloads with Kevin Hannon
Jay Faulkner sits down with Kevin (GitHub: canon92), Kubernetes Contributor Award winner and maintainer of the JobSet project, to discuss the growing challenge of running AI and batch workloads on Kubernetes — a platform originally designed for web services and long-running applications.
Kevin shares his journey from G Research and the Armada project through to his current work improving Kubernetes upstream, including the multi-year effort to bring swap memory support to GA, the push to deprecate CGroups v1, and the emerging Workload Aware Scheduling initiative aimed at bringing gang scheduling into Kubernetes core.
The conversation covers the technical realities of running GPU workloads at scale, the human side of open source maintenance, and the often overlooked work of keeping CI infrastructure healthy across a massive distributed project.
Topics discussed include JobSets, CGroups v2, pressure stall information, topology aware scheduling, and the open source politics of shipping features in a large community-driven project.
G-Research is hiring in Dallas, TX and London, UK! Apply at https://gresearch.com/vacancies.
For a video version of this podcast, check out https://youtu.be/7EQW57BlroA.
The GR-OSS OUT Podcast is produced by Ben Wiley.
Special Guest: Kevin Hannon.
View all episodes
By G-Research Open Source SoftwareJay Faulkner sits down with Kevin (GitHub: canon92), Kubernetes Contributor Award winner and maintainer of the JobSet project, to discuss the growing challenge of running AI and batch workloads on Kubernetes — a platform originally designed for web services and long-running applications.
Kevin shares his journey from G Research and the Armada project through to his current work improving Kubernetes upstream, including the multi-year effort to bring swap memory support to GA, the push to deprecate CGroups v1, and the emerging Workload Aware Scheduling initiative aimed at bringing gang scheduling into Kubernetes core.
The conversation covers the technical realities of running GPU workloads at scale, the human side of open source maintenance, and the often overlooked work of keeping CI infrastructure healthy across a massive distributed project.
Topics discussed include JobSets, CGroups v2, pressure stall information, topology aware scheduling, and the open source politics of shipping features in a large community-driven project.
G-Research is hiring in Dallas, TX and London, UK! Apply at https://gresearch.com/vacancies.
For a video version of this podcast, check out https://youtu.be/7EQW57BlroA.
The GR-OSS OUT Podcast is produced by Ben Wiley.
Special Guest: Kevin Hannon.