Managed clusters · Managed Slurm

Slurm, set up and kept running for you.

Get a Slurm cluster with partitions, a login node and shared storage, then submit work with sbatch or from the console. Supergate handles the setup and the upkeep.

A cluster your researchers already know.

A login node for SSH, partitions for GPU and CPU work, and a filesystem every node shares.

Managed Slurm cluster

Login node

  • ssh
  • sbatch

gpu partition

  • slurm
  • GPU × 8

cpu partition

  • slurm
  • 64 vCPU

Shared filesystem

  • /home + /data
Illustrative cluster with sample partitions and sizes.

What you get.

  • The scheduler they already use

    Researchers keep sbatch, srun and their scripts. Nothing new to learn before the first job.

  • Containerized jobs

    Submit containerized jobs to Slurm and follow each job as its state changes, from the console or the command line.

  • Kept healthy

    Supergate watches the cluster’s health signals and handles the upkeep, so a long run isn’t lost to a broken node. [TBD: auto-recovery]

Every job, from queued to done.

Search and filter jobs by type and status across the project, and open any one to see its metrics and logs.

  • Templates and scripts

    Keep job templates and scripts together and reuse them for the next run.

  • Files in and out

    Upload job inputs and download results without copying them through a login node.

  • Shared by many projects

    Project teams submit their own jobs, each with its own roles, on clusters they share.

Jobs screen listing jobs with their type, user, status and image.
Product screen with sample data.

Questions.

Do our existing Slurm scripts work?

Yes. Partitions, sbatch options and job scripts work as they do on any Slurm cluster.

Can jobs run in containers?

Yes. Submit a container image as a job, from the console or with your scripts.

How big can a cluster get?

Tell us the node counts and accelerators you need, and we’ll size the cluster with you. [TBD: published limits]

More solutions.