Skip to content

Scripts for launching and analysing parallel scaling with uw

Notifications You must be signed in to change notification settings

underworldcode/scaling_scripts

 
 

Repository files navigation

Scaling Scripts

This repo contains scripts to launch jobs for weak/strong scaling.

To launch a set of weak scaling jobs, you will generally modify the parameters in weak_params.sh as necessary, and then execute weak_scaling.sh which will queue a set of jobs on the system.

The executed jobs will generated JSON files which record timing information for the different execution stages. A equivalent user readable TXT file will also be generated which records a summary. This data is utilised by scaling_graphs.ipynb to generate graphs of the results.

timed_model.py

This is an underworld script which executes a simple model and records timing information. The model covers most of the fundamental underworld constructs, including solvers, vis and data IO.

weak_scaling.sh

This script will queue a set of jobs for testing underworld simulation weak scaling. It uses the settings in weak_params.py to determine run configuration.

weak_params.sh

This script contains run parameters for weak scaling tests. The user should generally only need to modify values here.

strong_scaling.sh

This script will queue a set of jobs for testing underworld simulation strong scaling. Note that this script is old and requires updating.

scaling_graphs.ipynb

This notebook utilises simulation results to generate scaling graphs.

magnus_container_go.sh

This script configures an environment and launches the required script on Magnus using containers.

magnus_baremetal_go.sh

This script configures an environment and launches the required script on Magnus running directly on machine.

gadi_baremetal_go.sh

This script configures an environment and launches the required script on Gadi running directly on machine.

About

Scripts for launching and analysing parallel scaling with uw

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 96.3%
  • Python 2.4%
  • Shell 1.3%