# Storage¶

The storage tutorial is a prerequisite.

These pages are also related and include solutions to common storage problems:

## Checklist¶

Do any of these apply to you? If so, consider your situation and ask us for help!

If you have been sent this checklist because your jobs may be having a lot of IO, don’t worry. It’s not necessarily a problem but please go through this checklist and let us know what applies to you so we can give some recommendations.

• Many small files being accessed in jobs (hundred or more).
• Files with extremely random access, in particular databases or database-like things (hdf5).
• Files being read over and over again. Alternatives: copy to local disks, or read once and store in memory.
• Number of files growing, for example all your runs have separate input files, output files, Slurm output files, and you have many runs.
• Constantly logging to certain files, writing to files from many parallel jobs at the same time.
• Reading from single files from many parallel jobs or threads at the same time.
• Is all your IO concatenated at one point, or spread out over the whole job?

( and if we’ve asked you specifically about your jobs, could you also describe what kind of job it is, the type of disk read and write happens, and in what kinds of pattern? Many small files, a few large ones, reading same files over and over, etc. How’s it spread out across jobs? )

If you think your IO may have bad patterns or even you just want to talk to make sure, let one of the Triton staff know or submit an issue in the issue tracker.

You can check the total disk read and write of your past jobs using:

# All your recent jobs:
sacct -o jobid%10,user%8,jobname%10,NodeList,MaxDiskRead,MaxDiskWrite -u $USER # A single jobid sacct -o jobid%10,user%8,jobname%10,NodeList,MaxDiskRead,MaxDiskWrite -j$jobid


(we will add more tools to this later)