Picongpu: tbg: environment before submit

Created on 17 Aug 2019  路  2Comments  路  Source: ComputationalRadiationPhysics/picongpu

First seen at NERSC: It's more efficient to prepare module and environment loading already on the head node instead of doing so in the first compute node of a job.

Potentially a workflow that should be covered by tbg, e.g. in its submitAction.sh.

tools refactoring

Most helpful comment

On slurm systems in general there is no need to load the environment systems again because slurm is storing loaded modules during sbatch.
We need to take care because slurm is not remembering all envirobment variables like the author for PIConGPU we set in our profile.

All 2 comments

On slurm systems in general there is no need to load the environment systems again because slurm is storing loaded modules during sbatch.
We need to take care because slurm is not remembering all envirobment variables like the author for PIConGPU we set in our profile.

Sounds good, so basically we can just simplify the profile template then to not load all modules on startup, and keep the simple bash variables such as author baked it.

Was this page helpful?
0 / 5 - 0 ratings

Related issues

berceanu picture berceanu  路  4Comments

steindev picture steindev  路  4Comments

cbontoiu picture cbontoiu  路  3Comments

ax3l picture ax3l  路  3Comments

sbastrakov picture sbastrakov  路  3Comments