SubmitFunction

class SubmitFunction(callable, timeout_per_job=5, ntasks_per_node=1, nodes_per_job=1, cpus_per_task=1, mem=512, slurm_wckey='P120K:SALOME', slurm_additional_parameters={})

The aim of this class is to ease the realization of parallel evaluations of a numerical simulation model in a HPC environment. This class gives an example of a HPC wrapper for an executable numerical model using the Python package submitit.

Parameters:

callableopenturns.Function

The unit function for which can either be sequential (a unit evaluation only requires one CPU), multi-cores or multi-nodes (a unit evaluation requires multiple cores and possibly multiple nodes).

timeout_per_jobint

Timeout requested (in minutes) per SLURM job.

ntasks_per_nodeint

Number of tasks (a task is a single evaluation of callable) that can be handled by a single SLURM job. Passed to SLURM as –ntasks-per-node.

nodes_per_jobint

Number of HPC nodes requested per SLURM job submitted. Passed to SLURM as –nodes.

cpus_per_taskint

Number of CPUs required to perform one task, i.e. one evaluation of callable. Passed to SLURM as –cpus-per-task.

memint

Memory (in MB) requested per node. Passed to SLURM as –mem.

slurm_wckeystr

Only for clusters that require a WCKEY (EDF clusters for example), i.e. a project identification key. To check the current wckeys, use the bash command cce_wckeys. Passed to SLURM as –wckey.

slurm_additional_parametersdictionary

Extra parameters to pass to SLURM (for example, {“exclusive”: True, “mem_per_cpu”: 12}). Empty by default.

Methods

getInputDescription()	Input description accessor.
getInputDimension()	Input dimension accessor.
getOutputDescription()	Output description accessor.
getOutputDimension()	Output dimension accessor.
setInputDescription(descIn)	Input description accessor.
setOutputDescription(descOut)	Output description accessor.
task(X)	Wrapper around callable to allow us to dispatch a single evaluation as a SLURM task

Examples

>>> import othpc
>>> import openturns as ot
>>> from othpc.example import CantileverBeam

>>> cb = CantileverBeam("my_results")
>>> slurm_cb = othpc.SubmitFunction(cb)
>>> X = [[30e3, 28e6, 250.0, 400.0], [20e3, 35e6, 250.0, 400.0]]
>>> Y = slurm_cb(X)

getInputDescription()

Input description accessor.

getInputDimension()

Input dimension accessor.

getOutputDescription()

Output description accessor.

getOutputDimension()

Output dimension accessor.

setInputDescription(descIn)

Input description accessor.

setOutputDescription(descOut)

Output description accessor.

task(X)

Wrapper around callable to allow us to dispatch a single evaluation as a SLURM task