Kubernetes with Kubeflow vs. Slurm: Scalability

Posted by ish-xyz on Friday July 05, 2024 @01:20PM
from the self-post dept.

Which one provides better scalability. Say I want to implement a 15k+ BM nodes HPC cluster... what would you go for?

( 4 comments | 3 points | on HN )

Re:Kubernetes with Kubeflow vs. Slurm: Scalability by pinewurst on Friday July 05, 2024 @02:26PM (#40884920)

Depends on the actual workflow and on the actual users. As a former manager of large HPC systems, I’d never ever burden myself with the Big K, as Slurm did a fine job for my workload and users.

[ Parent | on HN ]

Re:Kubernetes with Kubeflow vs. Slurm: Scalability by martialblog on Saturday July 06, 2024 @03:44AM (#40888822)

D'accord with that.

Slurm does the job well.

My users liked the Kubernetes API to throw their workloads against the clusters, but they did not need/use many on the Kubeflow features.

But again, depends on your workflow needs.

[ Parent | on HN ]

Re:Kubernetes with Kubeflow vs. Slurm: Scalability by higgsfield on Friday July 05, 2024 @04:22PM (#40885670)

slurm

[ Parent | on HN ]

Re:Kubernetes with Kubeflow vs. Slurm: Scalability by martialblog on Tuesday July 09, 2024 @02:18AM (#40912993)

Today I learned: SUNK, which stands for SlUrm oN Kubernetes

https://www.cncf.io/blog/2024/07/08/slurm-an-hpc-workload-ma...

[ Parent | on HN ]