Effect handlers benchmarks suite

The project aims to build a repository of systems that implement effect handlers, benchmarks implemented in those systems, and scripts to build the systems, run the benchmarks, and produce the results.

A system may either be a programming language that has native support for effect handlers, or a library that embeds effect handlers in another programming language.

Quick start

Ensure that Docker is installed on your system. Then,

$ make bench_ocaml

runs the OCaml benchmarks and produces benchmarks/ocaml/results.csv which contains the results of running the Multicore OCaml benchmarks.

System availability

System	Availability
Eff
Effekt
Handlers in Action
Koka
libhandler
libmpeff
Links
Multicore OCaml

Benchmark availability

Legend:

• :heavy_check_mark: : Benchmark is implemented
• :x: : Benchmark is not implemented
• :heavy_minus_sign: : Benchmark is unsuitable for this system, and there is no sense in implementing it (eg. benchmarking the speed of file transfer in a language that does not support networking)

Directory structure

• systems/<system_name>/Dockerfile is the Dockerfile in order to build the system.
• benchmarks/<system_name>/<benchmark_name>/ contains the source for the benchmark <benchmark_name> for the system <system_name>.
• descriptions/<benchmark_name>/ contains the description of the benchmark, the input and outputs, and any special considerations.
• Makefile is used to build the systems and benchmarks, and run the benchmarks. For each system, the Makefile has the following rules:
  • system_<system_name>: Builds the <system_name> docker image.
  • bench_<system_name>: Runs the benchmarks using the docker image for <system_name>.
  • test_<system_name>: Tests the benchmark programs using the docker image for <system_name>.
• LABELS.md contains a list of available benchmark labels. Each benchmark can be assigned multiple labels.

Contributing

Benchmarking chairs

The role of the benchmarking chairs is to curate the repository, monitor the quality of benchmarks, and to solicit new benchmarks and fixes to existing benchmarks. Each benchmarking chair serves two consecutive terms. Each term is 6 months.

The current co-chairs are

• Jesse Sigal (2023/09/27 - 2024/03/27 - 2024/09/27)
• Philipp Schuster (2022/09/21 - 2023/03/21 - 2023/09/21)

Past co-chairs

• Filip Koprivec (2022/01/21 - 2022/07/22 - 2023/03/21)
• Daniel Hillerström (Inaugural chair, 2021/07/23 - 2022/01/22 - 2022/09/20)

Benchmark

If you wish to implement <goat_benchmark> for system <awesome_system>,

• Add the benchmark sources under benchmarks/<awesome_system>/<goat_benchmark>. The benchmark takes its input as a command-line argument and prints its outputs.
• Update benchmarks/<awesome_system>/Makefileto build, test, and benchmark the program. Use the parameters for testing and benchmarking provided in descriptions/<goat_benchmark>/README.md.
• Update this README.md file to tick the new benchmark in the benchmark availability table.

Description

If you wish to add a new benchmark <goat_benchmark>,

• Add a benchmark description under descriptions/<goat_benchmark>/README.md. Use the template provided in descriptions/template/README.md.
• Provide a reference implementation for at least one system.
• Update this README.md and add a new row to the benchmark availability table.

System

If you wish to contribute a system <awesome_system>,

• Add a new dockerfile at systems/awesome_system/Dockerfile.
• Add a new workflow under .github/workflows/system_<awesome_system>.yml. It should build the system and run tests.
• Update this README.md and add a new column to the benchmark availability table. Create a status badge and add it as well.
• Update Makefile with commands that build, test, and benchmark the system.

Ideally, you will also add benchmarks to go with the new system.

Having a dockerfile aids reproducibility and ensures that we can build the system from scratch natively on a machine if needed. The benchmarking chair will push the image to Docker Hub so that systems are easily available for wider use.

We use Ubuntu 22.04 as the base image for building the systems and hyperfine to run the benchmarks.

Artifacts

We curate software artifacts from papers related to effect handlers. If you wish to contribute your artifacts, then please place your artifact as-is under a suitable directory in artifacts/.

There is no review process for artifacts (other than that they must be related to work on effect handlers). Whilst we do not enforce any standards on artifacts, we do recommend that artifacts conform with the artifacts evaluation packaging guidelines used by various programming language conferences.