Download your results
Everything a job produced lives on its Results page. This page covers getting it out, including the two cases the browser handles badly: multi-run jobs and sets over 4 GB.
The Results page
For a basic job, Results lists every file associated with the run.
Paths are ignored, and wildcards aren't supported. Searching log finds process_output.log anywhere in the tree; searching run1/*.log finds nothing.
View opens a file in the browser if it's under 1 MB. Larger files must be downloaded first.
Reuse outputs as inputs
Make selected files available as input files promotes an output into your input files, so a later job can start from it. This is how you:
- Chain a workflow — one job's output becomes the next job's input
- Restart from a checkpoint file a previous run wrote
- Rerun only the part of a study that failed
You can also clone the job directly with the selected files as its inputs, which combines this with duplicating.
Multi-run jobs
DOE and other multi-run job types produce many runs inside one job, each with its own inputs and outputs.
The all-runs zip includes each run's inputs and outputs. On a large sweep that's easily tens of gigabytes, and you probably wanted the summary or a few specific runs.
Start with the CSV. It gives per-run start/stop times and variable values, which is usually what you actually need to identify the interesting runs. Then download only those.
An individual run's detail view works exactly like the single-run Results page.
Files over 4 GB, or scripted downloads
Don't use the browser. For large sets or automation, use:
- Rescale Transfer Manager — parallel connections, resumes after a failure. The right tool for multi-gigabyte transfers.
- Rescale CLI — scripted and repeatable.
- REST API — programmatic access.
A browser download of a very large file has no resume: an interruption at 90% means starting over.
Downloading to a workstation instead
If you're going to post-process the results anyway, consider not downloading them at all. Attach the job to a workstation and work with the files in place, no transfer, and you get GPU visualization hardware. See move files in and out.
For a large result set this is usually faster than downloading, and it's the reason post-processing is the most common workstation use.
Storage keeps costing while files exist
Job files accrue storage charges until deleted. Once you have what you need locally, archive the job and empty the trash: archiving alone doesn't stop the bill.
Related
- Archive and delete jobs — stopping storage charges
- Rescale Transfer Manager — large transfers
- Move files in and out of a workstation — the no-download path
- Files and storage — the storage model
- DOE sweeps — where multi-run jobs come from