Aggregating Funnels

Software artifact associated with the following paper:

Aggregating Funnels for Faster Fetch&Add and Queues
Younghun Roh, Yuanhao Wei, Eric Ruppert, Panagiota Fatourou, Siddhartha Jayanti, Julian Shun
ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPoPP 2025)

Hardware/Software Requirements

• An Intel machine with 40+ cores (80+ hyperthreads) with Docker
• The full set of experiments takes ~10 hours to run

Getting Started Guide

1. Clone the repository

   • git clone https://github.com/Diuven/aggregating-funnels.git && cd aggregating-funnels

2. Build the docker image (install docker if you haven't)

   • docker build --network=host --platform linux/amd64 -t aggfunnel .

3. Launch the docker container as an interactive shell. This command also complies all the necessary binaries. Remaining commands should be run inside the docker container.

   • docker run -v .:/home/ubuntu/project -it --privileged --network=host aggfunnel
   • Note: This command mounts the current directory aggregating-funnels/ into the docker container, so both are synchronized.

4. Testing

   • Run python3 scripts/benchmarkRunner.py --task_path local/example.json inside the docker container
   • This should take a few minutes and produce an output graph in results

Reproducing Paper Results

5. Run the benchmark.

   • Running ./scripts/run_figures.sh will run all the experiments and generate the figures in the results/plots directory. This will take a couple of hours to run, depending on the number of threads available. stdout will show estimated time to completion for each figure.
   • During the benchmark, you can monitor the progress by checking stdout and the contents of the results/ directory. log.txt will have the output logs, and main.csv will have the raw data for each run.

Customizing the Benchmark

• The default workflow, as specified in scripts/run_figures.sh, is to run taskGenerator to generate task specs, run benchmarkRunner to run the tasks with the specified parameters, and finally run plotDrawer to generate the plots from the results.
  • taskGenerator generates the json file and saves to local directory. You can directly inspect and edit the json file (recommended), or run python3 scripts/taskGenerator.py to walk through the prompts and generate a custom task spec.
  • benchmarkRunner runs the tasks specified in the given json file. You can change the json file with --task_path option. It saves the results in the results directory, in a subdirectory specified by the json file's save_path field.
  • plotDrawer generates the plots from the results. It currently supports generating the plots for the figures in the paper. You can change --data_path, --save_path, and --figure_num options.

List of claims from the paper supported by the artifact

• Given a machine with > 80 logical cores, the graphs generated should be similar to the ones reported in our paper (up to the corresponding core count)