k8s-supporting-folding-at-home

k8s-supporting-folding-at-home copied to clipboard

Hi, Thx for publishing this. Have you thought on making this available in the helm chart public repo? Thx

Hi,

Yes, that is an idea for the future. I first wanted to get this chart to the public and more mature before creating a PR to the public repo :)

Last time a did a change to the public repo it took some months to get merged, so I created this repo first.

Let´s hope that we do not need this charts in some months.

We deployed this in a couple of clusters, and it seems to work, but not with GPU. If you have settings to switch between GPUs (nVidia/ATI/Intel) it would be great, as the chart could be even more plug-and-play. As for submitting chart, the merge can be quicker. In addition, we can add it to public chart repos, as with Helm3 things are a bit more distributed now.

https://chartmuseum.com/ has link to create free chart-registry account in codefresh (https://codefresh.io/codefresh-signup/)

Wow! That's really cool to hear!

https://github.com/wind0r/k8s-supporting-folding-at-home/issues/1 describes the changed needed to do to make it work at least with nVidia GPUs. I was able to successful test it with my Workstation but I do not have a GPU Cluster to validate the chart. Also, I still wanted to read about the License Agreements. :/

I will check how I can publish the helm chart :)

I think both things could be separate. There are already some docker images that seem to support GPU (https://hub.docker.com/r/coreweave/fah-gpu). Using this image should allow the chart to avoid licensing issues, and keep the focus on enabling the GPUs to be accessible on the pods, and aligned with the image.

Got this (the Docker image here: https://hub.docker.com/r/coreweave/fah-gpu ) running on a p3 machine on AWS. However, this requires some set-up on the machine: https://medium.com/@houdinisparks/deploying-gpu-containers-on-aws-ec2-with-docker-machine-c79c9062d90 And the use of --runtime=nvidia instead of the GPUs argument:

sudo docker run --runtime=nvidia --name fah -p 7396:7396 -v $(pwd)/fah-data:/root -d coreweave/fah-gpu:7.5.1

There is some information on how to get this setup in AWS here: https://aws.amazon.com/blogs/compute/running-gpu-accelerated-kubernetes-workloads-on-p3-and-p2-ec2-instances-with-amazon-eks/

Thx @fmannhardt. here some links on gpu on kubernetes and AKS

• https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/
• https://docs.microsoft.com/en-us/azure/aks/gpu-cluster
• https://github.com/Azure/aks-engine/blob/master/docs/topics/gpu.md