Skip to content

Latest commit

 

History

History
32 lines (20 loc) · 1.86 KB

README.md

File metadata and controls

32 lines (20 loc) · 1.86 KB
Raw

BROUTE

BROUTE is a benchmarking suite for routing optimization algorithms, see description here: https://arxiv.org/abs/2107.13492

General structure

All implementations are a subdirectory of the implementations directory. New implementations should also go there.

Each implementation directory contains a script called run_benchmark.sh which takes two arguments:

  1. the directory containing input data files to consider for the benchmark
  2. the benchmark name

For example to run the c++98 implementation on the lns benchmark using files from directory instances:

./implementations/c++98/run_benchmark.sh instances lns

Optionally, implementation subdirectories can contain a script called build.sh that executes necessary steps that are prerequisites for that implementation (e.g. compilation or package installation).

Running all benchmarks

Use scripts/build_all.sh to build all implementations.

Use scripts/run_all.sh, or scripts/complete_all.sh to generate only missing results. Both scripts look at the environment variable THREADS for parallelism. For example, to run all implementations of all benchmarks with 4 parallel threads: THREADS=4 ./scripts/run_all.sh

Exploiting benchmark results

There are scripts in the scripts directory for the analysis of experimental results:

  • sanitycheck.py takes CSV file names as command line parameters and checks that they all produce the same checksum values
  • compare.R produces box plots like in the paper. Requires R. A directory containing CSV files must be provided as command line parameter.
  • profiles.py produces performance profiles like in the paper. Requires matplotlib. A directory containing CSV files must be provided as command line parameter. The profiles are generated interactively: for each profile, the legend can be moved interactively then the PDF is generated when the window is closed.