Create and submit a job
The reference version of job setup. If this is your first job, the tutorial covers the same path with a worked example.
If you or a colleague has run something similar, duplicate it or launch from a template and skip this page.
The five steps
Setup is a five-step wizard, shown down the left of the page. You can move between steps freely before submitting.
Inputs → Software → Hardware → Automations → Review
The right-hand Job Details rail stays visible throughout, holding the job's Project, Job Type, Tags, description, and custom fields. You can collapse it to give the setup steps more room.
Starting a job
From Jobs, click New Job. A drawer opens with two ways in:
The template list is searchable and shows each template's coretype, core count, software and version, wall time, tags, and who created it. Scan it before starting blank: see reuse a job or template.
Name the job in the field at the top of the page. You'll be scanning a list later, so make it identifiable.
If your workspace uses CFS, the job name becomes part of the filesystem path and can't be changed afterwards. Worth a second's thought before you continue.
1. Inputs
Attach the files your run needs. Three sources, with different cost and speed:
Add From Cloud Storage also reaches connected external storage (S3, Azure Blob, SharePoint) under External Storage. Those files transfer straight to the cluster at launch and add no Rescale storage charge.
You'll also see Output Filters on this step, which limit what gets saved back: see post-processing.
Set the Job Type in the Job Details rail. Basic runs your command once; other types run it many times.
If your workspace uses CFS, the job name becomes part of the filesystem path and can't be changed afterwards. Worth a second's thought at step 1.
Use compressed archives
Rescale automatically extracts .zip, .tar, .tar.gz, .tgz, and .gz into the working directory before the run starts. Uploading an archive is faster to transfer, cheaper to store, and the archive stays available for reuse.
Contents are extracted into the working directory, so paths in your command are relative to what's inside the archive, not to the archive. If your command expects to run inside case01/, either compress at that level or prepend a cd:
cd case01 && run_analysis
Getting this wrong is the most common cause of "file not found" on a first job. See troubleshooting.
2. Software
Search the catalog for your application and select it. Then set:
- Version. Pin it if your results need to be reproducible; solver versions change behavior.
- Command. What Rescale actually runs, executed from the working directory.
Separate multiple commands with a newline, ;, or &&:
Use && when a later step is meaningless if an earlier one failed.
You can call the solver directly or run a script you uploaded as an input file. A script is the right choice when the run is a sequence: preprocessing, several solver invocations, then post-processing.
Command Templates
Software Settings offers Command Templates: suggested command lines for the standard or serial form of an application, for SMP, and for MPI. They exist to save you assembling one from scratch.
Some solvers also ship a Rescale wrapper script (ls-dyna, for example) that
simplifies the call. Using the wrapper is optional; the underlying executable is on
$PATH either way.
Replace every placeholder before you submit. Templates use angle brackets for the parts only you know:
ls-dyna -i <input-file> -p <precision>
Substitute a real value and delete the brackets with it. If a field doesn't apply to your run (a hybrid flag when you only want MPI), remove it entirely rather than leaving it empty.
For the command each solver expects, see Solver run commands.
3. Hardware
Three fields for a basic job:
Walltime is a failsafe, not a reservation. It exists to hard-stop a job that would otherwise run indefinitely. You are billed for time used, so a generous walltime costs nothing, while a job that hits its walltime is stopped mid-run and loses unsaved progress.
Below the coretype, a table shows what you're actually getting: Nodes, Cores/node, Cores, Memory/node, Memory/core, Storage/node, GPUs/node. Check the memory-per-core figure against what your solver needs. It's the most common cause of a job that dies partway.
Watch for the two cost prompts
The Hourly Price Summary in the left rail shows the hourly rate for the hardware you have selected, and updates as you change coretype or core count. It is a rate, not a projected total: the platform does not estimate what a job will cost, and walltime does not affect the figure.
If your coretype belongs to a Coretype Collection, a prompt appears offering to "use the Collection for equal or better performance at the same cost" with a Switch to Collection button.
Taking it lets Rescale pick any equivalent-or-better coretype in the collection, which improves your odds of getting capacity without costing more. See coretype collections.
The coretype also carries a priority badge (On Demand Priority, Economy, or Reserved) which trades cost against interruption risk. Your workspace sets the default. See job priority.
4. Automations
Optional post-processing that runs on the cluster after your command finishes. Use it to extract, reduce, or plot results before they transfer, rather than downloading everything and post-processing locally.
Skip this step for a simple run. See automations for what you can script.
5. Review and submit
Check the summary, then Submit. Save stores the configuration as a template without running it.
For Coretype Collection jobs, the Review page shows the Collection name before submission. After the job completes, it displays Collection → Actual Coretype (for example, "Emerald Plus → Emerald Max"), showing which specific coretype ran.
Rescale checks that your workspace can cover the first hour of the job, or has a valid card on file. If not, the job fails its precondition and never launches.
This is a first-hour check, not a whole-job check. A long job can pass submission and then be stopped mid-run if funds run out. See why a job is queued and budget enforcement.
Configure licensing if needed
For commercial solvers, either use Rescale's on-demand licensing or bring your own.
To use your own: check Use Existing License and enter your license server, typically as port@hostname. Rescale remembers it, so later jobs pick it from a dropdown.
You can also declare license feature usage, which lets Rescale hold the job in the queue until the features you need are actually free rather than starting it and failing on checkout. See licensing and why a job is queued.
Post-processing (optional)
Two things you can do:
- Filter output files so only what you want is saved back — the Output Filters control on the Inputs step.
- Run a post-processing script on the cluster via Automations, so data is extracted and reduced before transfer rather than after download.
Add custom fields (optional)
Custom fields attach structured business data to the run. A CAD part number, a load case, a drag coefficient. Your workspace may require some before you can submit; those show a red asterisk.
Click Add Custom Fields in the Job Details rail to add job- or template-level fields. Optional fields can be filled after the run finishes, which is how findings get recorded.
Full detail (scopes, the eight field types, and the API) is in custom fields.
Submit from the CLI or API
For chained jobs, scheduled batches, and pipelines, submit outside the UI. A CLI submission uploads inputs, points at a job configuration JSON, and polls for completion.
- Chain jobs with the CLI — submit a job from inside a running job
- Submit with High-Performance Storage attached via API
See REST API and developer tools.
Troubleshooting
Submit is greyed out. A required field is missing, usually no software version, or an empty command. See the greyed-out submit button.
The job is stuck in Queued. It's waiting on a precondition, budget, capacity, or licenses. See why a job is queued.
Job failed immediately at Validating Input. Your input files or command didn't check out. See troubleshooting.
Related
- Run your first job — these steps with a worked example
- Reuse a job setup — duplicates and templates
- Job priority — the cost/interruption tradeoff
- Monitoring & status — after you submit
- Troubleshooting — when it doesn't work