About SDSU OnDemand
SDSU OnDemand is an HPC web-based portal built with our University’s research community in mind. OnDemand integrates with our cluster, Roaring Thunder, and offers features such as job submission, job monitoring, shell access, and interactive graphical sessions and apps all from within a web browser. SDSU OnDemand is built from Open OnDemand, an NSF-funded open-source project maintained and developed by the Ohio Supercomputer Center (OSC).
OnDemand is supported on all operating systems and browsers. Although not officially supported at this time, OnDemand is accessible via mobile as well.
Logging In
To access OnDemand, visit ondemand.sdstate.edu and log in with the same username and password you use to log in to the cluster. You must have access to the cluster in order to use SDSU OnDemand. Please contact us if you do not have access to the cluster and would like to be added to the access group.
Using OnDemand
OnDemand’s Files menu allows access to your /home directory on RT (and your /scratch directory, if you have one). The web-based file explorer allows you to remotely interact with your files on RT. You can create files and folders, view files, change file locations, upload files, and download files. You can also show dotfiles (hidden files) and file permissions.
The Active Jobs menu shows all jobs you currently have running on RT and their status (completed, running). You can view your own jobs, or all jobs running on the cluster, using the dropdowns to filter your search.
The Job Composer is made up of two parts: Jobs and Templates. The Jobs menu allows you to create a new job to submit to the cluster, either from the system’s default Slurm template, another available template, a specified path on the filesystem, or from a previously selected job.
Using the default template as an example, we are able to edit a Slurm script to submit to the cluster. On the right side of the screen, you can customize your script by clicking “Open Editor” and adding SBATCH options, modules you’d like to load, and job instructions. You can open the directory your job script is stored in to add additional files used for your job, and you can also open your job in a terminal on RT.
Once your job is ready to submit, you can click “submit” and your job will be submitted to the cluster. You can watch the status of your job by clicking on the Active Jobs menu described in the previous step. You can also stop your job or delete your job.
There are example templates included with OnDemand as well. To access example templates, click the “New Job” dropdown as shown in the above image and click From Template. There are several job templates available to give you a general idea of how these Slurm jobs are run on the cluster. You can also create your own templates, available only to you, by clicking “New Template” or “Copy Template”. These are stored in:
/home/$USER/ondemand/data/sys/myjobs/templates/
Each one of these system templates can be run on the cluster as-is. To create a job from a template, select one and click Create New Job. You will be brought back to the Jobs menu where you can edit these job files or submit them to see how they work.
If there is a system template you would like us to create or add for you or your group, please contact us.
Please note: when submitting jobs via the Job Composer, OnDemand will create a special job directory for you in your /home directory:
/home/$USER/ondemand/data/sys/myjobs/projects/default/1 …
This path cannot be changed at this time, however, there will be an option to change this path in an upcoming release of OnDemand which we plan to integrate. If you prefer, when your job is finished, your files can be moved elsewhere in your home directory by using the File Explorer.
The Clusters menu allows you to open a web-based terminal that connects you through an SSH session to the login node on RT. OnDemand offers SSH access to RT within your browser. No SSH client is required to access the cluster.
Interactive Applications
There are Interactive Apps available for use on the cluster with OnDemand. The interactive apps menu provides direct web access to interactive applications on the cluster’s worker nodes without the need to edit job scripts.
We will use JupyterLab as an example.
When selecting JupyterLab as an interactive app, you are able to choose options for launching your JupyterLab application on the cluster. You can select compute, bigmem, or GPU for the type of node you would like your application to run on. There are other options such as the number of cores, hours, and nodes for your job. Finally, you can select the CUDA version should you choose to use a GPU node.
Note: When selecting options for your job, please follow the guidelines on the Interactive Apps page for maximum cores, nodes, etc. Any erroneous entries may cause your session to fail to start.
Below is an example of an Interactive App form on OnDemand.
Once you have launched your application, your session will be set up on the cluster. Depending on how busy the cluster is and how many resources you request, your session may be in queue for several minutes until the cluster can accommodate your request.
You will see this window as your application loads:
And when it is ready:
Select "Connect to Jupyter" to begin your session.
More applications are being developed for use with OnDemand. In a later release of OnDemand, we plan to integrate an update that allows users to start an interactive desktop on the cluster.