Direct comparison
srun vs sbatch vs salloc: Slurm Commands
Compare srun, sbatch, and salloc: when to submit a batch job, run interactively, or allocate resources for interactive HPC work on Slurm.
Written and maintained by CASRAI Editorial Board
Last updated
Ask CASRAI · included with Regulatory Radar
Ask about srun vs sbatch vs salloc: Slurm Commands
Ask CASRAI answers research-administration questions and cites the passages behind every claim — and says so when the corpus does not cover something, instead of guessing. It comes with a Regulatory Radar subscription at $29 a month, alongside the daily digest of regulatory changes and the dashboard of what changed.
150 questions a day, on this site, over the API, or inside your own tools through the CASRAI MCP server.
Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.
How do sbatch, srun, salloc compare side by side?
The table below compares sbatch, srun, salloc across 8 procurement-relevant dimensions, from what it does through common flags.
Side-by-side comparison
| Dimension | sbatch | srun | salloc |
|---|---|---|---|
| What it does | Submits a batch script to the Slurm scheduler for non-interactive execution later, when resources are available. | Runs a command directly under Slurm — either as its own job (standalone) or as a parallel job step inside an existing allocation. | Requests a resource allocation and starts an interactive shell (or runs a specified command) within it. |
| Execution mode | Non-interactive, asynchronous. | Interactive/blocking — waits for the command to finish and streams output to your terminal. | Interactive — you get a live shell running inside the allocation. |
| Returns control | Immediately, with a job ID. Your terminal is free right away. | Only after the launched command finishes. | As soon as the allocation is granted, dropping you into a shell. |
| Typical use case | Production runs: long simulations, pipelines, anything unattended. | Launching parallel/MPI tasks inside a batch script, or a quick one-off interactive command. | Interactive debugging and development directly on compute resources before writing a batch script. |
| Can it run inside an existing allocation? | No — sbatch always requests a new job allocation; it cannot be nested inside salloc or another sbatch job. | Yes — this is its dual role. Standalone it creates its own allocation; inside sbatch/salloc it runs as a job step using the parent allocation's resources. | N/A — salloc creates the allocation that srun commands then run inside. |
| Output handling | Stdout/stderr go to a file (slurm-<jobid>.out by default) unless --output/--error is set. | Streams directly to the terminal (standalone) or to the batch script's own output stream (inside sbatch). | Whatever you run inside the shell behaves normally — output goes to your terminal. |
| Ends when… | The script finishes (or hits its time limit / is cancelled). | The launched command completes. | You exit the shell (or run scancel), releasing the allocation. |
| Common flags | --job-name, --partition, --time, --nodes, --ntasks, --cpus-per-task, --mem, --output, --array | --ntasks, --cpus-per-task, --nodes, --pty, --exact | --nodes, --ntasks, --time, --partition |
Common questions
Common questions about sbatch vs srun vs salloc
Can I run srun without sbatch or salloc first?
+
Yes. Run standalone, srun requests its own resource allocation from Slurm and then executes the command, blocking your terminal until it finishes. That's fine for a quick interactive test, but for a session with several commands, salloc is usually the better starting point because it keeps the allocation open.
What does srun actually do inside an sbatch script?
+
It launches a job step that runs within the allocation sbatch already obtained — it doesn't request new resources. By default a step launched this way can use all the non-GRES resources in that allocation unless you constrain it with flags like --exact, --ntasks, or --cpus-per-task.
Does salloc run my program for me?
+
No. salloc only creates the allocation and starts an interactive shell (or runs a specified command once, if you give it one). To actually launch work across the allocated nodes and tasks, you still call srun from inside that shell.
Which one should I use for interactive debugging on a cluster?
+
salloc, for a session where you'll run several commands, or srun --pty bash for a single one-off interactive shell. sbatch is the wrong tool for debugging — it's built for unattended runs where nobody is watching the output live.
Why does my sbatch job show up as one job in the queue no matter how many srun steps it runs?
+
Because sbatch creates a single job allocation, and every srun call inside its script runs as a step within that same job — steps share the job ID (shown as <jobid>.<stepid>), they do not create separate queued jobs.
Do the default resource limits differ between the three commands?
+
Slurm doesn't fix universal default values — time limits, memory per CPU, and partition access are set by each cluster's own configuration. Check your cluster's documentation, or run scontrol show partition, rather than assuming a default that applies everywhere.
