Problem-Based Benchmark Suite in Elixir

Goal

Rewrite the Problem-Based Benchmark Suite in Elixir.

To run

Install and configure Elixir 1.14+, Erlang 24+ on your machine. Alternatively, you can use Docker to build an image with all the dependencies and the code, as defined in Dockerfile. For example (extracted from run_benchmarks.sh):

docker build . -t elixir-pbbs
image_id=$(docker images | awk '{print $3}' | awk 'NR==2')

container_id=$(docker run -d -it "$image_id" /bin/bash)
docker exec -it "$container_id" sh -c "mix benchmark -a histogram"

Benchmarks are executed with the mix benchmark task. To execute a benchmark:

mix benchmark -a <algorithm>

Currently implemented algorithms:

• histogram - https://cmuparlay.github.io/pbbsbench/benchmarks/histogram.html
• word_count - https://cmuparlay.github.io/pbbsbench/benchmarks/wordCounts.html
• remove_duplicates- no PBBS page :(
• ray_cast - https://cmuparlay.github.io/pbbsbench/benchmarks/rayCast.html
• convex_hull - https://cmuparlay.github.io/pbbsbench/benchmarks/convexHull.html
• suffix_array - https://cmuparlay.github.io/pbbsbench/benchmarks/suffixArray.html
• integer_sort - https://cmuparlay.github.io/pbbsbench/benchmarks/integerSort.html
• comparison_sort - https://cmuparlay.github.io/pbbsbench/benchmarks/comparisonSort.html

Each benchmark has a number of different, preselected inputs, to exercise different communication patterns in the parallel implementations. Each benchmark compares the parallel implementation with the sequential implementation for a given problem, given the same inputs. Each implementation-input pair executes for 60 seconds, plus 2s for warmup.

Benchmarks output a CSV file with detailed runtime information to the current directory. Currently, the benchmarks are not parameterizable, so feel free to change the code in utils/benchmarks/drivers/* if necessary.

License

MIT