Same sbatch, new system of record- Matt Bonyak | Domino

HPC runs the biggest jobs but often keeps the fewest records. A simulation runs for three days, its output lands in results-final-v2, and six months later nobody can trace the inputs, code, or environment behind the reported results. As AI workloads move onto the same clusters, that gap becomes more costly.

In this session, we'll show how Domino adds a system of record to Slurm without changing existing workflows. With real Slurm managed by Slinky, scripts still run through sbatch, squeue, and sacct. Domino Workspaces serve as login nodes, where coding agents like Claude Code or Cursor can write and submit jobs, with prompts routed through the LLM Gateway and logged alongside each job. Domino captures requested and actual resources, environment and code revisions, file versions, and NetApp data snapshots at execution time.

We'll demonstrate an unchanged Slurm script, workers autoscaling across x86 or ARM tiers and back to zero, and one-click reproduction of a months-old run—including its original environment and data.