Overview

To start a job:

  1. Go to the “Job” tab.
  2. Fill out the form to specify the resources you want for the job.
  3. Check the graphs at the bottom of the page to be sure that the cores and memory you are requesting are available. If you request more resources than are currently available, it may take hours or days for your job to start. Reduce what you are requesting, if necessary, to ensure that your job starts right away.
  4. Click the “Start Job” button at the bottom of the page. It may take several minutes for your job to start — if it doesn’t start within 10 minutes, please contact hpc-support@umich.edu for assistance.

connect-starting

A status window will show you the progress of your job. “Queued” means that your job is waiting to run. “Starting” means that your job has started running and is setting up the ARC Connect session you requested. You can use the “Cancel” button at any time to delete your job.

If you told your web browser to allow ARC Connect to send you notifications, you will receive a notification as soon as your job finishes starting and is ready to use.

What happens when your job finishes starting depends on the session type you chose when you filled out the Start Job form:

VNC:

  • A new tab containing your desktop will open automatically.
  • To end the job, click the green “running person” icon at the bottom of the desktop.
  • Switch back to the Job tab to get VNC connection information for using a desktop VNC client or for sharing the session with collaborators. You can also forcefully terminate the job from the Job tab.

Jupyter or RStudio:

  • Click the green “Open In Browser” button to access your Jupyter or RStudio session.
  • Be sure to save your work before using the red “Terminate” button on the Job tab to end the job.

Details

Resource: Click the button for the HPC cluster where you want to run a job. Currently, only Flux is supported; Armis will be added as soon as ARC Connect has been reviewed for HIPAA suitability.

Account: Select the resource account / allocation(s) you want to use to run the job.

Session type: Choose the type of job you want to run:

  • VNC: A desktop environment with windows that you can use to run graphical software interactively. You can share your desktop with collaborators, allowing them to view it, or to control it. The desktop can be accessed either through your web browser or via [link: using a VNC viewer application].
  • Jupyter: Formerly known as IPython, Jupyter allows you to create scientific notebooks containing a mix of text, code, and results. The code can be Python 2, Python 3, R, or Julia.
  • RStudio: RStudio is an integrated development environment (IDE) for R. (R is a computer language for statistical computing and graphics.) RStudio includes a console, syntax-highlighting editor that supports direct code execution, as well as tools for plotting, history, debugging and workspace management.

Desktop resolution: This option controls the size of the VNC desktop. This option only appears if you selected “VNC” as the Session type.  Most people will want to select a resolution that is lower than the resolution of the laptop, desktop, or other device from which they are accessing ARC Connect. You can request a VNC desktop that is the same or higher resolution than your computer or tablet, but you will only see part of the desktop at a time and will have to scroll in order to see other areas of the desktop.

Generic Resource (gres): If you know that you need to request a generic resource for your job, such as a specific software license, select it here. If you need to select something here, you will receive a notice when you load the software module, and/or it will be documented on one of the ARC-TS web pages. Leave the value at “(none)” unless you know you need a generic resource.  This option only appears if you selected “VNC” as the Session type.

Generic Resource (gres) quantity: How many of the generic resource your job requires. To avoid creating problems for other researchers, you should request only what you need. For software licenses, some software requires 1 gres per job, while other software requires 1 per node, process, or core; consult the ARC-TS documentation for the software you are using to determine what is required in your specific case. This option only appears if you selected “VNC” as the Session type. Additional information is available on the Requesting Software Licenses web page.

Nodes: How many nodes (computers) your job requires. You can request multiple cores on each node. If you select more than one node, the software you run inside your job needs to be capable for utilizing multiple nodes (via MPI or some other mechanism). As you increase the number of nodes, you may find that the choices for number of cores per node decrease.

GPUs: This option only appears if you select a resource account which has GPUs (Graphical Processing Unit cards, sometimes also know as GPGPUs, General Purpose Graphical Processing Units). You should leave this at the default value, 1, unless you are running software which is capable of using multiple GPUs simultaneously. For additional information, refer to the GPGPU web page.

Cores: How many cores your job requires on each node.  Increasing cores per node will increase performance of software that uses either multiple threads per process and/or multiple processes (including software which uses MPI), until the limit of the software’s parallelism is reached. The maximum number of cores you can request is based on the size and type of the resource account you selected, as well as the number of nodes you are requesting. If you select more cores than are currently available (unused) in the resource account, a warning will be displayed telling you that the job may take a very long time to start.

Memory: How much memory you want per core, in gigabytes. Be sure to request 20% more memory than the maximum you think your job needs in order to avoid having your job terminated by the cluster resource manager for using too much memory. When you end an ARC Connect job, you will receive email saying how much memory your job used so that you can adjust your memory request for future jobs. If you select more memory than is currently available (unused) in the resource account, a warning will be displayed telling you that the job may take a very long time to start.

Walltime: How long you want the job to run. The job will be terminated by the cluster resource manager when it reaches the requested walltime. You can either click in each field (days, hours, minutes, seconds) to type a number, or you can use the up/down buttons.

Start Job: Click the “Start Job” button to submit your job after filling out the form.

Clear choices: Click the “Clear choices” button to reset all of the fields above to their default values.

Resource utilization:  Shows core availability, memory availability and a list of jobs  in the currently selected resource account.