The SLAC Shared Science Data Facility (S3DF) is a compute, storage and network facility supporting SLAC science.
Documentation is available here.
For getting help and reporting problems email s3df-help@slac.stanford.edu.
For general discussion see the #comp-sdf channel on the SLAC Slack workspace.
If you are a SLAC employee or affiliated researcher (i.e. with KIPAC) with a SLAC computing account, you are eligible for an S3DF account. To enable your S3DF account, log into coact using your SLAC computing credentials and follow the instructions to request membership in the KIPAC S3DF facility.
(Note that S3DF is presently in a transition from Unix to SSO/ActiveDirectory authentication for all S3DF web and SSH bastions anticipated to be complete in the Fall of 2025. Unix account passwords will not work with S3DF services after the transition. If you currently only have a Unix account you can provision your SSO/ActiveDirectory account via a self-service portal at https://ad-account.slac.stanford.edu using your Unix account and follow the directions to create your SSO/ActiveDirectory account.)
There are 3 clusters (–partitions in Slurm terminology) at S3DF: Roma, Milano, and Ampere (GPU). KIPAC facility S3DF members have priority access (non-preemptable jobs) on 20 Roma nodes and 4 Milano large memory nodes. Each Roma node has 128 AMD EPYC 7702 cores and 512 GB RAM. KIPAC facility S3DF members can also run low-priority preemptable jobs on Milano and the GPU cluster Ampere.
Directories in /sdf/data/kipac/u are intended for the exclusive use of each corresponding user. Please only write to the directory that corresponds to your username, and keep all files and subdirectories group-readable so we can track usage within the group. The current default limit is 1TB for each user. This limit may increase in the future as we increase capacity and/or free up space. If you need more than the current limit, please contact Marcelo Alvarez with your request so we can coordinate and increase your quota if necessary.
No compute tasks should be done on the login nodes since those are meant to operate only as bastion hosts, not for doing analysis or accessing data. Instead, use the “iana” pool of interactive nodes. A typical login session would be:
% ssh username@s3dflogin.slac.stanford.edu
% ssh iana
Should you need dedicated resources for an interactive job, you can use, e.g.
srun --account kipac:kipac -n 1 --time=01:00:00 --mem-per-cpu=16G --pty /bin/bash
Note the <facility>:<repo> repo pattern (where both are kipac) which should help reduce queue times and prevent preemption, whether the jobs are interactive or not. See here for more details.