The vast majority of the calculations we run are carried out on supercomputers, including those at Princeton. You should bookmark the Princeton Research Computing Knowledge Base , which has many excellent guides for using the high-performance computing (HPC) clusters. The most important rule is to treat the research computing staff with respect. Abide by the rules, when reporting issues do so with a complete and reproducible example, and be both kind and thankful. They are our colleagues, and you should treat them as such even if you have not met them before in person. Anyone found being rude or disrespectful to a computing staff member will have their compute access temporarily revoked.
The command line is how you will interact with the various compute clusters and run calculations. If you are on Linux, you already have a native command line. If you are on a Mac, you also have a native command line, although there are some promising new options like warp . If you are on Windows, I recommend using MobaXterm . The Princeton Research Computing webpage has a guide for logging into the HPC clusters via SSH:
Please set up SSH keys for increased security and convenience. For some clusters, there are also web browser options:
To access and/or transfer files on an HPC machine, the easiest option is to install an SFTP client. For Windows, I strongly recommend WinSCP . Additional options can be found below: In general, you will likely find it convenient to mount the remote drives to your local machine so that you can use your machine's built-in file explorer:
See the link below on editing code remotely on the clusters using VS Code.
Slurm is the job scheduling system on the Princeton HPC machines and most clusters we use. Details about Slurm can be found in this guide by Princeton research computing. The Slurm documentation also has a useful list of commands. To run a lightweight process in the background on the login node (e.g. to submit jobs), you can use tmux to preserve the process even when the SSH session is terminated.
Running tmux new -s launcher will create a new tmux session named launcher. To exit the tmux session while still preserving any running jobs on the login node, press ctrl+b followed by d. To re-enter thetmuxsession, run tmux attach -t launcher. Additional tmux commands can be found on the tmux cheatsheet . To view and interact with GUI applications on the cluster, refer to the instructions below:
There can be a fair amount of tedious things with HPC clusters. Check out Princeton's Removing Tedium guide to make your life a bit easier.