Princeton Research Computing has an excellent website that can address many preliminary questions you may have.
Description of Resources
A list of the various compute resources we currently use in the group is shown below:
Tiger: A restricted use cluster that we have access to, mainly meant for parallel jobs that use at least one full node. Serial calculations and calculations using less than a full node should not be run on Tiger unless multiple are packed within a single job. Tiger is especially useful for running "wide" jobs where many CPU nodes are used.
At the time of writing, we have contributed 13 CPU nodes (2x Intel 8480s w/ 112 cores per node and 1 TB memory) to Tiger3, which increases our priority.
We also have a dedicated login node at <NetID>@tiger-arrk.princeton.edu
Della: A general-purpose, mixed CPU/GPU cluster. You should mainly use this for GPU codes of any size and for CPU calculations using less than a full node (e.g. serial jobs).
Neuronic: A small GPU cluster for members of SEAS.
Adroit: A cluster meant for debugging and testing purposes. If you're developing new software and running simple tests, this can be a useful resource to do rapid prototyping. Production calculations are never run here. Adroit is where our production tests for quacc are run.
If there are issues with Princeton computing resources, you should submit a ticket. Always be courteous and detailed with your questions or requests. The IT staff are our friends, and we want to keep it that way.
File Systems
To view file quotas, use checkquota. If we are in danger of running low, inform me as soon as possible so we can take prompt action.
Home Directories
The /home directory on Adroit, Della, and Tiger is limited in terms of file storage but is backed up by Research Computing. In general, /home should be used sparingly. It is much slower than /scratch/gpfs.
You have a /home/<NetID> directory as well as a shared one for the group at /home/ROSENGROUP.
Scratch Directories
The /scratch/gpfs (Della, Tiger) or /scratch/network (Adroit) directories are where you should do must of your day-to-day work. It is fast NFS-mount disk space with a fairly generous amount of storage space.
You have a /scratch/<NetID> directory as well as a shared one for the group at /scratch/gpfs/ROSENGROUP (Della, Tiger) or /scratch/network/ROSENGROUP (Adroit).
The /scratch/gpfs space is not backed up. You must have a data storage plan that involves regular, scheduled backups of your data to a more permanent location (see below).
Data Backups
This section is currently a work in progress, as Research Computing is refining their data management protocols. Please talk to me about how to perform data backups in the meantime, and let's update this section when we have further info.
General Comment
If you ever feel resource-constrained in the group, let Andrew know so that the problem can be resolved. In addition to national computing allocations mentioned above, the group also has dedicated funds that can be used to contribute to the Princeton HPC resources, thereby increasing our priority in the queue.