Lightning Talk: Managing Drivers in a Kubernetes Cluster
As a cluster operator, managing drivers (Mellanox networking, NVIDIA compute and graphics drivers, …) at scale today is a real issue, from installation to upgrade every step you take brings you furt …
| Talk Title | Lightning Talk: Managing Drivers in a Kubernetes Cluster |
| Speakers | Renaud Gaubert (Software Engineer, NVIDIA) |
| Conference | KubeCon + CloudNativeCon Europe |
| Conf Tag | |
| Location | Barcelona, Spain |
| Date | May 19-23, 2019 |
| URL | Talk Page |
| Slides | Talk Slides |
| Video | |
As a cluster operator, managing drivers (Mellanox networking, NVIDIA compute and graphics drivers, …) at scale today is a real issue, from installation to upgrade every step you take brings you further away from Kubernetes. Drivers are frequently needed for enabling users (e.g: run AI workloads) or reducing cost (RDMA over converged ethernet), yet there are no clear consensus or tools that allows you to solve the issues encountered by requiring drivers on your machines. During this Lightning talk we’ll take a look at the different strategies you can use in Kubernetes to manage drivers (containers vs base image) and the available update strategies that will help you minimize disruption and maximize cost. Finally we will take a look at the challenges and solutions that VM based runtimes introduce.
