Build and Run an Experiment
Glossary terms: Benchmark, Experiment, System
Before initializing the system, open experiments/hpl/experiment.py and configure the
exec_mode=test branch to use one MPI rank and a problem size of 1,000:
if self.spec.satisfies("exec_mode=test"):
self.add_experiment_variable("n_ranks", 1, True)
self.add_experiment_variable("Ns", 1000, True)
Initialize the AWS tutorial system and the HPL experiment. The explicit instance type matches the Benchpark tutorial infrastructure:
$ benchpark system init --dest=aws-tutorial aws-tutorial instance_type=c7i.12xlarge
$ benchpark experiment init --dest=hpl aws-tutorial hpl+strong caliper=time,mpi
Create the experiment workspace:
$ benchpark setup aws-tutorial/hpl workspace
The setup.sh script generated by benchpark setup activates Benchpark’s Spack and
Ramble environments:
$ . workspace/setup.sh
Set up the Ramble workspace to build HPL and generate each experiment instance, then run the experiments:
$ ramble --workspace-dir workspace/aws-tutorial/hpl/workspace workspace setup
$ ramble --workspace-dir workspace/aws-tutorial/hpl/workspace on
For workspace structure and additional build options, see Build an Experiment. For individual experiment instances, rerunning experiments, and combining workspaces into one allocation, see Run an Experiment.