Cloud Native Innsbruck September

Join us as at Data Lab Hell in Zirl, for two talks:

CNCF Ambassador Thomas Schütz Lessons learned from not (really) using Kubernetes

How often did you read a post claiming that Kubernetes is too complicated? But what would a world without Kubernetes actually look like? Over the last few months, I started experimenting and tried to see how that world looks! In this talk, I will share my insights from projects built without Kubernetes and all the fancy things we know from the CNCF landscape. You will leave this talk hopefully with a smile, some ideas on when it might be a good choice to skip Kubernetes, what the alternatives look like, and a fresh appreciation for kubectl when you return to your office tomorrow.

Daniel Baldauf Many Tools, Many Models: Sharing GPUs with Ray on Kubernetes

Ray, Flyte, MLflow, a package index and Kubernetes underneath. Each of them solves one part of the job and someone has to make them work together. This talk shows the concept we settled on. Ray sits in the middle. KubeRay runs the worker pods, several workloads can share one GPU and workers shut down when nobody is using them. Flyte starts the training runs, MLflow keeps the results and Ray Serve answers requests. A new model is a versioned package and a few lines of configuration. One approach covers different kinds of tasks and models instead of a separate setup for each. I will show how that works in practice and why we built it this way.

Agenda

17:30 - Open Doors

18:00 - Talks & Discussion

~19:30 - Food, Drinks & Good Conversations!

After the talks, take the opportunity to network with peers and iscuss the latest trends and technologies in the cloud-native ecosystem, share experiences, and build new connections.

We're looking forward to seeing you for an evening of fun and learning!