This repository will provide the code that accompanies the paper by Guozheng Zhang, Gilead Posluns, and Mark C. Jeffrey, presented at the 36th ACM Symposium on Parallelism in Algorithms and Architectures (SPAA), June 2024.
This is a super repo for submodule benchmarks, the Multi Bucket Queue (MBQ) & the MultiQueue (MQ) (both in the cps repo).
The MBQ and MQ are implemented separately, and are embedded into the individual submodule frameworks for evaluation and comparison against the native scheduler in that framework.
# cloning the main repo
git clone https://github.com/mcj-group/mbq.git
# getting all the submodule repos recursively
cd mbq
git submodule update --init --recursive
Each framework contains benchmarks used for evaluation, some contain native schedulers which we use for comparison against our MBQ and MQ.
All benchmark repos are minimalistic (only include necessary files for experiments).
| Framework | Benchmarks | Input Type | Native Scheduler |
|---|---|---|---|
| galois | sssp, ppsp, bfs, pr |
.gr galois binaries |
OBIM, PMOD |
| ligra | sc |
PBBS adjacency graphs | Julienne |
| pbbs | mis |
symmetric PBBS adjacency graphs | Synchronous Parallel Loops |
| rbp | rbp |
specified & generated at run-time | |
| astar | astar |
.bin graph binaries with coordinates |
cps repo: contains the MBQ & MQ (and others, see README from cps's main branch), which are necessary for comparing with above benchmarks. The schedulers in this repo require Boost.
See the README.md file in each individual frameworks/repos for more details on prerequisites and info.
All the repo have scripts for building and running experiments.
We do not provide input files (see download sources from the paper).