Repo containing code for benchmarking SCIP performance on a PBS-TORQUE cluster.
The benchmarking is executed using atools, a library for easily running parameter sweep on an HPC cluster.
- Generate configurations to run with benchmark_setup.py -a [size|n_workers]: Test scaling with respect to increasing dataset size or number of workers. This creates a directory containing a results folder and data.csv file containing the configurations.
- Run configurations with an array job.
- Gather all runtimes in a timing-results.csv file using benchmark_post.py.
An attempt was made to run the configurations with Snakemake as it also allows to dynamiccaly request resources based on what configuration is run (for example, setting ppn based on the number of workers). It almost works, but still sometimes the execution fails due to a missing vsc-mympirun implementation.