Using OpenOnDemand

Open OnDemand

Open OnDemand is a portal for interactive access to MSI’s compute clusters, accessed through your web browser. Through its interface, you will be able to access your MSI files, a display of your current jobs on the MSI clusters, a command line on either of our compute clusters, or an interactive desktop on a compute node of either cluster with configurable memory and compute resources. In addition, you will have access to familiar servers, interactive development environments, and graphical user interfaces that have been adapted to use the Open OnDemand platform.

Note that if you are not already connecting from on-campus, you will need to connect to the University VPN before you can access Open OnDemand. To use Open OnDemand, point your web browser at https://ondemand.msi.umn.edu



   



In your browser window, the menu bar will look something like this, with several pull down menus, and other important functions.

Logout of your Open OnDemand session (button on far right): When you are finished with your interactive session, please Logout. See below.

Interactive Apps: Here you can start up a desktop instance on a compute node of one of our clusters, or use one of the applications we have configured to use Open OnDemand. We will start with a limited number of popular options here, and may add more as the service matures and we get requests. The initial Servers will be for Rstudio and Jupyter.

Rstudio server is an integrated development environment for R, a free software programming language for statistical computing and graphics. For more information on R, see https://www.r-project.org/

Jupyter is a free, open-source, interactive web tool known as a computational notebook. The default on MSI's OnDemand installation is with a Python notebook. Here are some OnDemand detailed examples for doing deep learning with Pytorch, and creating custom Jupyter notebooks with Singularity.

Desktop is an application that allows you to start up a Linux virtual desktop directly on a compute node, within which you are able to access MSI software resources and run command line tools. The resources you specified when you started the session (compute cores, memory) are available to you for the time you have requested. There are pulldown menus for you to select which MSI cluster you want to run on, which of your accounts (if you have more than one) to use, the compute and memory resources you wish to use, and the time limit of your interactive Linux session.

Example: You want to use an application that makes use of a GPU and does some 3-D graphics, which some current users of MSI’s NICE utility do now. You could select the Agate cluster, and the “Interactive GPU” entry under the resources menu, and then “4 Hours” under the time limit menu.

Example: Your interactive job does not need a GPU or a large amount of resources, but it needs to run for a long period of time, which describes how some current users of MSI’s NX-NoMachine utility use the system. You could select the Mesabi cluster, and the “Interactive Long” entry under the resources menu, and then “24 hours” (or “custom” to specify a longer time limit) under the time limit menu.

For these examples, and for all of the applications you can start under the Interactive Apps menu, you will have some choices to make regarding the compute and memory resources you will need to run your application. We suggest starting first with a small number of compute cores and memory, and increase it when/if you discover your work needs the extra resources.

If you would like to use a dedicated node that your group has leased for your interactive application, you will need to manually enter your private partition name, by choosing "Custom" under the resources menu, and "Custom" under the partitions menu, then type in the name of your private partition.

Files: Here you will find the directories/folders you have in your primary storage. There is also a clickable user interface to upload and download files to/from your connecting machine.

Jobs: A list of your active jobs on the clusters.

Clusters: Here you can open a command line window on either Agate or Mesabi login nodes (i.e., not a compute node). [Mangi is part of the Mesabi cluster.]

Other menu bar buttons allow you to look at a list of your active sessions, restart your Open OnDemand server, and most importantly, Logout of your Open OnDemand session. Always close your active sessions and Logout of Open OnDemand when you are done doing your interactive computing. There is a small limit to the number of simultaneous interactive sessions you can have open, so if you do not Logout of a session you are done with, it will stay open and perhaps impede you from later opening a new session.

Acknowledgements

Open OnDemand was originally developed by Ohio Supercomputer Center supported by two NSF grants, and is under active development by a team from the Ohio Supercomputer Center, the University of Buffalo, and Virginia Tech University. https://openondemand.org/