Platform R: Getting Started

Before You Log In

See Platform R: Account Activation for getting a Platform R account.

Connecting to Platform R 

1. Head over to Azure Virtual Desktop (AVD).
   
   • Sign in using your NetID@wisc.edu email.
     • Note: If you are signed out, it will redirect you to the UW-Madison SSO sign-in page to complete the sign-in process.
       
2. You will see a Desktop labeled "Platform R PHI Desktop".
   
   
   • Click "Connect".
3. Set your session settings.
   
   • Check "Clipboard" but uncheck "Keyboard shortcuts." The shortcuts can interfere with cut-and-paste.
4. Enter your NetID credentials when prompted by AVD to "Sign In"
   
5. Use PuTTY, SecureCRT, or VSCode to finish connecting to the Platform R Cluster.
   • Using PuTTY as an example:
     • Connect to host: slurm.platformr.wisc.edu port: 22
        
     • Save and cache the server's SSH key fingerprint by clicking "Accept".
       
     • Use you NetID credentials to sign in when prompted.
       
       • Note: Using PuTTY, you can save your username to auto-populate:
         
     • This host, slurm.platformr.wisc.edu, is only for logging in, moving data, setting up your environment for workflows, and submitting slurm jobs. It should never be used for computation.

See Platform R: Azure Virtual Desktop for more details about the Platform R Azure Virtual Desktop environment.

Ingress/Egress of Data for Platform R

Research and Restricted drive are approved methods of ingress of data for Platform R, however Restricted Drive is the only approved egress of data to Platform R. Simply, users have Read/Write permissions to Restricted Drive, but only Read permissions to Research Drive.

Note: The movement of data can only be performed from the head node within Platform R (pr-sett-01). 

1. List the contents of Restricted Drive (do this first - sometimes Restricted Drive will not be connected without it).

   • ls -ll /mnt/restricteddrive/NETID

2. Copy the sample data folder and it's contents to the group scratch location - by default, files in the destination folder will be overwritten upon transfer. 

   • rsync [options] [source file/directory] [destination]

     • example: rsync -av --progress /mnt/restricteddrive/NETID/sampledata /mnt/scratch/group/NETID/

3. Do your work.
4. Copy your results back to the Processed folder within Restricted Drive.

   • rsync -av --progress /mnt/scratch/group/NETID/sampledata /mnt/restricteddrive/NETID/processed 

Useful tips about "rsync":

• -a (or -archive) enables archive mode - this creates a recursive copy of the data.
• -u (or -update) will skip files that are newer in the destination location.
• --ignore-existing will skip updating files that already exist in the destination folder.
• A trailing slash (/) will alter the behavior of the transfer.

 Looking for further information about Platform R: Data Ingress/Egress or rsync?

Using Slurm for Job Management

Slurm is workload manager and job scheduling software utilized in Platform R, used for checking availability of resources, queuing and running jobs, and checking the status of in progress jobs, among many other features. 

See the clusters resources

• Check the state of cluster resources from the head node (pr-sett-001):

  • sinfo

Run a job

• Utilize the "srun" command to run your job with specific parameters.

  • srun --cpus-per-task=32 --mem=256G --partition=gpu --gres=gpu:nvidia_h200:4 --pty my_script.sh

    • Note: "gres" is Slurm's allocation of GPUs, and "my_script.sh" is the script you will be running.
• Once your job is running, you can check on the reservation of the GPUs by using "nvidia-smi"

  • nvidia-smi

• Check your queued jobs by using "squeue"

  • squeue -u <NetID>

Looking for further Slurm documentation? Check out Platform R: Slurm Introduction.

Logging Out

Return to the head node by entering exit in the command prompt and pressing enter. Then, by entering exit again, your method of connection will close out. All Slurm jobs submitted will continue to run and return output without you needing to be connected, this includes giving up the allocated resources dedicated to the job.