Once you land on the login nodes, either via a terminal or via a NoMachine desktop, you will need to ssh to one of the interactive pools to access the data, build/debug your code, run simple analyses, or submit jobs to the batch system. If your organization has acquired dedicated resources for the interactive pools, use them; otherwise, connect to the S3DF shared interactive pool.
?> Note: After log in into our bastion hosts with ssh s3dflogin.slac.stanford.edu, you will need to then also log into our interactive nodes to access batch compute and data. You can do this via ssh <pool name> within your ssh session (same terminal) to get into the bastion hosts.
The currently available pools are shown in the table below. (The facility can be any organization, program, project, or group that interfaces with S3DF to acquire resources.)
| Pool name | Facility | Resources |
|---|---|---|
| iana | For all S3DF users | 4 servers, 40 HT cores and 384 GB per server |
| rubin-devl | Rubin | 4 servers, 128 cores and 512 GB per server |
| psana | LCLS | 4 servers, 40 HT cores and 384 GB per server |
| fermi-devl | Fermi | 1 server, 64 HT cores and 512 GB per server |
| faders | FADERS | 1 server, 128 HT cores and 512 GB per server |
| ldmx | LDMX | 1 server, 128 HT cores and 512 GB per server |
| ad | AD | 3 servers, 128 HT cores and 512 GB per server |
| epptheory | EPPTheory | 2 servers, 128 HT cores and 512 GB per server |
| cdms | SuperCDMS | (points to iana) |
| suncat | SUNCAT | (points to iana) |
| neutrino | Neutrino | (points to iana) |
| mli | MLI (ML Initiative) | (points to iana) |
Under some circumstances, for example if you need more, or different, resources than available in the interactive pools, you may want to run an interactive session using resources from the batch system. This can be achieved through the Slurm command srun:
srun --partition <partitionname> --account <accountname> -n 1 --time=01:00:00 --pty /bin/bash
This will execute /bin/bash on a (scheduled) server in the Slurm partition <partitionname> (see partition names), allocating a single CPU for one hour, charging the time to account <accountname> (you'll have to get this from whoever gave you access to S3DF), and launching a pseudo terminal (pty) where bash will run. See batch banking to understand how your organization is charged (computing time, not money) when you use the batch system.
Note that when you 'exit' the interactive session, it will relinquish the resources for someone else to use. This also means that if your terminal is disconnected (you turn your laptop off, loose network etc), then the job will also terminate (similar to ssh).
To support X11, add the "--x11" option:
srun --x11 --partition <partitionname> --account <accountname> -n 1 --time=01:00:00 --pty /bin/bash
Users can also access S3DF through Open OnDemand via any (modern) browser. This solution is recommended for users who want to run Jupyter notebooks, or don't want to learn SLURM, or don't want to download a terminal or the NoMachine remote desktop on their system. After login, you can select which Jupyter image to run and which hardware resources to use (partition name and number of hours/cpu-cores/memory/gpu-cores). The partition can be the name of an interactive pool or the name of a SLURM partition. You can choose an interactive pool as partition if you want a long-running session requiring sporadic resources; otherwise slect a SLURM partition. Note that no GPUs are currently available on the interactive pools.
After login onto OnDemand, select the Clusters tab and then select the interactive pool from the pull down menu. This will allow you to obtain a shell on the interactive pools without using a terminal.
We provide automatic tunnels through our ondemand proxy of Jupyter instances. This means that in order to run Jupyter kernels on S3DF, you do not need to setup a chain of SSH tunnels in order to show the Jupyter web instance.
We also provide the capability for you to 'bring-your-own-Jupyter' so that all your code dependencies are dictated by you, and not by us. We recommend you do this by either building an apptainer image of your Jupyter environment or by building a conda environment on SDF storage.
If you wish for your jupyter environment to be more widely used (e.g. for others in your group), you can submit a pull-request to our slac-ood-jupyter repo to append your specific "Commands to initiate Jupyter" onto the list of preselectable Jupyter Images. Specifically, you will want to change form.yml.erb.
Once you have created your Conda environment on SDF (see Software/Conda), ensure that you have jupyter installed in your conda environment:
conda install -c conda-forge jupyterlab
Then go to Jupyter portal, and select "Custom Conda Environment..." from the "Jupyter Instance" dropdown. You will need to customize the text that appears under "Commands to initiate Jupyter" to point to your custom conda environment:
export CONDA_PREFIX=<path-to-miniconda3>
export PATH=${CONDA_PREFIX}/bin/:$PATH
source ${CONDA_PREFIX}/etc/profile.d/conda.sh
conda env list
conda activate <your-environment-name>Replace <path-to-miniconda3> and <your-environment-name> appropriately.
Fill the rest of the form as you would for any provided Jupyter Instance and click "Launch". If you run into any issues, please see Debugging your interactive session.
Once you have built or pulled an Apptainer image on SDF (see Software/Apptainer page for more information on how to do that), ensuring that you have the jupyter[lab] binary in the image's PATH, go to the Jupyter portal, select "Custom Apptainer Image" from the "Jupyter Instance" dropown menu. Then modify the text in the "Commands to initiate Jupyter":
export APPTAINER_IMAGE_PATH=<path-to-the.sif>
function jupyter() { apptainer exec --nv -B /sdf,/lscratch ${APPTAINER_IMAGE_PATH} jupyter $@; }Replace <path-to-the.sif> with the full path to your local apptainer image file.
Fill the rest of the form as you would for any provided Jupyter Instance and click "Launch". If you run into any issues, please see Debugging your interactive session.
If you get an error while using your Jupyter instance, go to the My Interactive sessions page, identify the session you want to debu and click on the Session ID link. You can then View the output.log file to troubleshoot.