singularity.rst

Singularity for Satorians

Fast start

• Pull your image

export SINGULARITY_DOCKER_ARCHITECTURE=ppc64le
singularity pull docker://centos:7

• Get an allocation, for example:

srun --gres=gpu:4 -N 1 --mem=1T --time 1:00:00 -l --pty /bin/bash

• Submit your payload

singularity exec centos_7.sif ls /

Other notes

• Pull your image
• If you are using an existing image, confirm that arch is ppc64le

Interactive Allocation:

• Get an allocation (for interactive allocation):
• Consider the size of your workload, do you need CPU? memory? GPU?
• These are all specified in the bsub request, here is a good [1]
• For example, here is a bsub to request allocation for 4 GPUS, via the normal queue, launching an interactive session using bash as terminal

srun --gres=gpu:4 -N 1 --mem=1T --time 1:00:00 -l --pty /bin/bash

• Submit your payload
• Using interactive allocation, we are allocated a node - your interactions with singularity will be generally standard , just like you would in any environments
• Remember you do not have sudo access on this cluster, this might impact your workflow

singularity exec centos_7.sif ls /

Non interactive / batch mode

• Workflow is similar to interactive allocation, however we need to create an application profile prior to submitting the workload
• This mode might be better for headless / non-interactive workloads as the resources can be more easily allocated to the workload
• Create an application profile [2]
• For example

Begin Application
NAME = singapp
CONTAINER = singularity[image(centos_7.sif)]
DESCRIPTION = Singularity User Service
End Application

• Submit the workload via the application profile [3]
• For example

sbatch singapp ./myjob.sh

[1]	(https://www.ibm.com/support/knowledgecenter/SSWRJV_10.1.0/lsf_command_ref/bsub.man_top.1.html)

[2]	(https://www.ibm.com/support/knowledgecenter/en/SSWRJV_10.1.0/lsf_container/lsf_singularity_config.html).

[3]	(https://www.ibm.com/support/knowledgecenter/en/SSWRJV_10.1.0/lsf_container/lsf_singularity_use.html).