Resource documentation feedback

[consideRatio]

Please use this GitHub issue to provide feedback on our documented resource requests!

Please accompany feedback with information about...

• Helm chart version
• Estimated simultaneous users during heavy load
• Estimated maximum users spawning a server over a 15 minute duration

[brospars]

Hello,

I'm deploying Jupyterhub on k8s for a machine learning course. We expect 10 000 users in total and about 200 daily users.

We used to have a monolithic setup based on https://github.com/jupyterhub/jupyterhub-deploy-docker :

• Config : 128 CPU / 512 Go RAM
• Max simultaneous users : ~500 over 15min

The new k8s setup :

• Config :
  • Nodepool for JH (1 node : 2 CPU / 7 GO RAM)
  • Nodepool for Users (10-50 nodes : 4 CPU / 15 Go RAM)
• Max simultaneous users : ~150 over 15min then JH crashes (503 and 429 errors)

I'm using a modified version of yuvipanda/hubtraf to run heavy machine learning code (requesting CPU and RAM)

The aim is to have a scalable setup that can reach 500 simultaneous users. For now I'm unable to reach that goal but I'll update this post when I find the bottleneck (it's probably due to few CPUs on the JH node).