Getting Started with RIS Compute

Problem

You need to get onto the RIS compute cluster to perform some task.

Solution

Log into an RIS jumpbox:

ssh -X -Y -l <your-wustl-login-id> compute1-client-1.ris.wustl.edu

Then invoke a bsub command:

LSF_DOCKER_VOLUMES="/home/$USER:/home/$USER /storage1/fs1/I2/Active:/storage1/fs1/I2/Active /scratch1/fs1/I2:/scratch1/fs1/I2" \
LSF_DOCKER_PORTS="8201:9999" \
LSF_DOCKER_PRESERVE_ENVIRONMENT=true \
  bsub \
  -Is \
  -M 30GB \
  -n 4 \
  -m compute1-exec-131.ris.wustl.edu \
  -R 'rusage[mem=30GB,tmp=30] select[gpuhost && mem>32GB && tmp>30] span[hosts=1]' \
  -G compute-ohids-t2 \
  -q ohids-interactive \
  -a 'docker(jupyter/scipy-notebook)' \
  /bin/bash -l /opt/conda/bin/jupyter notebook --port 8201

Discussion

Coming soon!

We need to ensure that we say the following:

-sla will not work when -q is ohids-interactive or ohids
if -q is one of the above please use -R 'gpuhost'
hosts have GPUs they are being tagged as gpuhost. And hence, to use them, jobs need to have that argument. Users are completely free to use them in any way possible (CPU only, GPU only, or mix of both). But in every case the argument is required in all bsub jobs that wish to run on ohids condo
"If your condo/queue contains gpus, like the ohids-interactive or ohids queues, then the -R 'gpuhost' must be specified for all bsub commands, whether your job uses GPUs or not."

References

• RIS Cluster Documentation
• RIS Cluster Documentation: Job Execution Examples
• Condo and Subscription LSF Queues
• Docker Wrapper Environment Variables