Published November 9, 2026 | Version v1

A Parallel Greedy Algorithm for Maximal $b$-Matching in Hypergraphs - Code & Experimental Data

  • 1. ROR icon Wayne State University
  • 2. ROR icon Heidelberg University

Description

We present PB-bM, a parallel batch-vote algorithm for maximal
$b$-matching in hypergraphs. 
The algorithm is deterministic for a given ordering of edges and
produces the same output as the sequential greedy algorithm, i.e., the lexicographically
first maximal $b$-matching. Under a uniformly random priority order it performs
$\mathcal{O}(m')$ expected work in the binary fork--join model, where $m'$ is the
number of pins, the total number of vertex--edge incidences.  

The algorithm performs at most $m$ rounds for every priority order, where $m$ is the number of edges.
Every round of the algorithm has $\mathcal{O}(\log^2 m')$ depth with high probability, and thus the total depth of the algorithm is $\mathcal{O}(m \log^2 m')$ with high probability.
We implement the parallel algorithm in CUDA and evaluate it on real-world hypergraphs with capacities $b \in
\{1, 3, 5\}$. The parallel algorithm computes the same matchings as the greedy algorithms 
and for the case of $b = 5$ it is up to $19.07$ times faster than the existing state-of-the art greedy algorithm.

Please cite 

```

@misc{bakhshan2026parallelgreedyalgorithmmaximal,
      title={A Parallel Greedy Algorithm for Maximal b-Matching in Hypergraphs}, 
      author={Saeed Bakhshan and Daniel Grosu and Henrik Reinstädtler and Christian Schulz},
      year={2026},
      eprint={2611.15615},
      archivePrefix={arXiv},
      primaryClass={cs.DS},
      url={https://arxiv.org/abs/2611.15615}, 
}

```

or the paper to appear in ACDA'27

## Artifact Information

This repository contains the code and experimental data used in the experiments.

Parallel-main.zip: Mirror of the published source code at https://github.com/HeiHGM/Parallel

results.zip: The raw data used in the paper.

experiments.zip : Files needed to run the experiments to validate the plots in the paper

instances.zip: Instances file

### Instruction to reproduce.

unzip the instances.zip file, we reference its data path as $INSTANCE_PATH below

Unzip experiments.zip. Run for each capacity the following command (replacing <1,3,5> by one of it):

export CUDA_MODULE_LOADING=EAGER # This is important because otherwise CUDA will lazy load code
./bazel-bin/runner/kokkos_runner --experiment_path=/path/to/folder/containing/final/<1,3,5>  --random_order 1 --max_alloc_memory   --max_alloc_memory_per_process   --cycles_to_queue_new 1 --seed 1234 --root_path=$INSTANCE_PATH

Please choose sensible values for and in MB. So something like 50000 and 100000.

Files

experiments.zip

Files (11.0 GB)

Name Size
md5:4a3ed7134d153fdb706b0bbd6fa02c53
15.1 kB Preview Download
md5:784ad42d46f5cca4e2a04ad99a321c76
11.0 GB Preview Download
md5:441c8ea24e53632aebb99a6a96532e29
189.5 kB Preview Download
md5:2d8c7796b3bdfdb7ed248ffad6ff8a9b
715.4 kB Preview Download

Additional details

Funding

Deutsche Forschungsgemeinschaft
543787284

Software

Repository URL
https://github.com/HeiHGM/Parallel
Programming language
C++