Five ways to get software
- System modules (recommended). Check first:
module avail software-name, thenmodule load software-name/version. - Conda environments. Best for Python and data science packages such as numpy, pandas, TensorFlow and PyTorch. See Conda environments.
- Python venv. Best for lightweight Python projects without compiled dependencies.
- Install from source. Use this when the software isn't a module or a package, or when you need to build it your own way. See Installing from source.
- Request a system-wide install. Submit at iacs.supportsystem.com for multi-user needs.
Where to install
| Location | Quota | Backed up | Persistent | Best for |
|---|---|---|---|---|
Home /gpfs/home/ | 20 GB | Yes | Yes | Personal tools, small apps |
Project /gpfs/projects/ | Large | No | Yes | Shared tools, group use |
Scratch /gpfs/scratch/ | 20 TB | No | No (45-day purge) | Temp files ONLY |
Important: don't install software in scratch. Files there are deleted after 45 days, even if you still use them, and your software will stop working.
Best practices
- Keep each tool in a folder named for what it does and its version, for example under
$HOME/software/. - Check your quota often with
myquota, and clean the conda cache withconda clean --all. - Write down how you installed each tool and what it needs.
- Set group permissions with
chmodon software you share.
Applies to